> ## Documentation Index > Fetch the complete documentation index at: https://docs.tensor9.com/llms.txt > Use this file to discover all available pages before exploring further. # Bedrock (LLM inference) costs > Compare Bedrock (LLM inference) backend choices using explicit service-adapter planning assumptions. Tensor9 BYOC planning comparisons for `aws::1.0.0::bedrock`, with backend choices shown together. Read [how Tensor9 estimates service-adapter costs](/byoc/costs/overview.md) for price scope and capacity qualification, or return to [additional services](/byoc/costs/additional-services.md). ## Bedrock model inference → Anthropic Messages API Origin API: `aws::1.0.0::bedrock`. Target backend: `anthropic::1.0.0::messages`. ### Bedrock model inference · Input tokens per second Estimated monthly service charges by Input tokens per second. Adapter hosting and Tensor9 fees are excluded. Source: us-east-1; target: global; currency: USD. ![Bedrock model inference · Input tokens per second: monthly service charges by Input tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/bf25808fe4922a0ad0c18074441a5fd8169c51a6bacc4a10811bca373aab2f91-light.svg) ![Bedrock model inference · Input tokens per second: monthly service charges by Input tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/a25e1fd318f23959e202c53f98184f13b5f842e766cd863c79d85f37e50bd6a6-dark.svg) Comparison pills describe all sampled scale points; card amounts show the named workload. Costs The Same means within 20% at every point. Costs Depend includes mixed results and unpriced samples. Exact sampled model outputs in USD/month. Difference means target minus origin; a positive difference means the target costs more. | Input tokens per second (tokens/s) | Bedrock (LLM inference) | Target scenario | Anthropic Messages API | Difference | | -: | -: | - | -: | -: | | 0 | 375 | Anthropic Messages API | 375 | 0 | | 10 | 453.84 | Anthropic Messages API | 453.84 | 0 | | 100 | 1163.4 | Anthropic Messages API | 1163.4 | 0 | | 1000 | 8259 | Anthropic Messages API | 8259 | 0 | | 10000 | 79215 | Anthropic Messages API | 79215 | 0 | | 100000 | 788775 | Anthropic Messages API | 788775 | 0 | **Curve assumptions** | Target scenario | Assumption | Value | Unit | | - | - | -: | - | | Anthropic Messages API | Origin list price · Input tokens | 0.000003 | USD/Tokens | | Anthropic Messages API | Origin list price · Output tokens | 0.000015 | USD/Tokens | | Anthropic Messages API | Target / origin input tokens | 1 | multiplier | | Anthropic Messages API | Target / origin output tokens | 1 | multiplier | | Anthropic Messages API | Target price guess · read | 0 | USD/read | | Anthropic Messages API | Target price guess · write | 0 | USD/write | | Anthropic Messages API | Target price guess · provider operation | 0 | USD/provider operation | | Anthropic Messages API | Target price guess · GiB-month | 0 | USD/GiB-month | | Anthropic Messages API | Target price guess · published GiB | 0 | USD/published GiB | | Anthropic Messages API | Target price guess · delivered GiB | 0 | USD/delivered GiB | | Anthropic Messages API | Target price guess · network GiB | 0 | USD/network GiB | | Anthropic Messages API | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | Anthropic Messages API | Target price guess · vCPU-hour | 0 | USD/vCPU-hour | | Anthropic Messages API | Target price guess · GiB-hour | 0 | USD/GiB-hour | | Anthropic Messages API | Target price guess · resource-hour | 0 | USD/resource-hour | | Anthropic Messages API | Target price guess · resource-month | 0 | USD/resource-month | | Anthropic Messages API | Target list price · input token | 0.000003 | USD/input token | | Anthropic Messages API | Target list price · output token | 0.000015 | USD/output token | | Anthropic Messages API | Target monthly base charge | 0 | USD/month | | Anthropic Messages API | Approximate target service minimum | 0 | USD/month | | Anthropic Messages API | Minimum primary CPU | 2 | vCPU | | Anthropic Messages API | Minimum primary RAM | 8 | GiB | | Anthropic Messages API | Minimum primary disk | 100 | GiB | | Anthropic Messages API | Additional full-price standby copy | 0 | 0 or 1 | | Anthropic Messages API | Allocated share of provisioned cost | 1 | fraction | | Anthropic Messages API | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | Anthropic Messages API | Planning peak / mean work | 2 | multiplier | | Anthropic Messages API | Planning primary CPU utilization | 0.7 | fraction | | Anthropic Messages API | Resident state reserved in RAM | 0.1 | fraction | | Anthropic Messages API | Planning disk utilization | 0.8 | fraction | | Anthropic Messages API | Maintenance and log disk reserve | 20 | GiB | | Anthropic Messages API | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | Anthropic Messages API | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | Anthropic Messages API | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | Anthropic Messages API | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | Anthropic Messages API | Dependent object sink write price guess | 0.000005 | USD/write | * Only this native meter varies; all other source usage and translation assumptions retain the displayed defaults. Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. Other native meters retain their explicitly displayed default quantities. * Planning estimate · editable authored rates and sizing assumptions dated 2026-10-22; not verified provider prices. us-east-1 → global, USD/month. * Source prices are reference charges, not an invoice. Target lines cover selected service usage/configuration only; adapter hosting and Tensor9 fees are excluded. * Explicit tariff samples are not performance measurements or capacity qualification. No interpolation or break-even guarantee is implied. * Variant assumptions show the first axis sample; any axis-dependent configuration is described by this curve's workload note. * Rates are cited provider list prices, editable for discounts, commitments or another region. Workload translation and capacity coefficients remain editable planning assumptions, not benchmark measurements or a commercial quote. * Default-zero rates are explicit assumptions, not evidence that the deployed service or resource is free. * Provisioned-unit minimums use an editable approximate allocation floor. Modeled target service cost is max(usage/allocation, minimum), with billable network, durable staging and dependent sinks added separately; actual RU/TU sizing, scaling increments and throughput remain unqualified. * Durable staging is priced once: provisioned targets include adapter state in their selected allocation; native targets add a separate staging allocation when required. Adapter hosting is a separate allocation and must exclude resources already priced as staging, workers or target capacity. * A declared adapter mapping and cost model do not establish implementation availability or service equivalence. * This is a declared directed mapping; a planning estimate does not establish deployment or request-serving support. * Origin unit prices are cited provider list prices; capacity coefficients and translation assumptions are editable authored planning guesses, not measured capacity or a provider quote. * Origin tariffs are the cited Claude Sonnet 4.6 global-endpoint list prices; target tariffs name no asserted current model SKU. Select model-specific rates before using the estimate. * Input and output meters and token expansion remain independent. A cheaper model is not automatically equivalent in capability or quality. * Input and output token tariffs remain distinct and establish no equivalent output quality or tokenizer. Claude Sonnet 4.6, the Claude model the Bedrock adapter's evals and the Vertex Claude arm default to. inference\_geo 'us' applies 1.1x; cache hits (0.30/MTok) and batch (50% off) are not modeled; Claude Sonnet 5 lists 2/10 per MTok. * AWS Price List API, AmazonBedrockFoundationModels offer: [https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json) * Claude API pricing: [https://platform.claude.com/docs/en/about-claude/pricing](https://platform.claude.com/docs/en/about-claude/pricing) Cost scope: Modeled target native service and explicitly modeled state allocation; hosting and Tensor9 fees only when supplied. Excluded costs and workload features: * Prompt caching, batch discounts, context tiers, grounding, tools, agents, embeddings, search, guardrails and dedicated inference hardware. * Taxes, discounts, credits and free-tier allowances unless explicitly reflected in edited rates. * Unmodeled dependent services, task execution and operations outside the selected workload family. List prices read: 2026-10-22. Model: `plan-ccde7cd673441e8ba203e027d2e5f895e072eb7c` / `2026-10-22.planning.2`. Curve: `workload-1`. Sources: [AWS Price List API, AmazonBedrockFoundationModels offer](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json), [Claude API pricing](https://platform.claude.com/docs/en/about-claude/pricing). ### Bedrock model inference · Output tokens per second Estimated monthly service charges by Output tokens per second. Adapter hosting and Tensor9 fees are excluded. Source: us-east-1; target: global; currency: USD. ![Bedrock model inference · Output tokens per second: monthly service charges by Output tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/87c77a173ae1be7d554db085ab7b6abaf72cfdd4910a557f121d3cc0e6727c04-light.svg) ![Bedrock model inference · Output tokens per second: monthly service charges by Output tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/65fbc1333f11453d94f92b78e190e46a92243f36c23db91b71586d04cf7d2074-dark.svg) Comparison pills describe all sampled scale points; card amounts show the named workload. Costs The Same means within 20% at every point. Costs Depend includes mixed results and unpriced samples. Exact sampled model outputs in USD/month. Difference means target minus origin; a positive difference means the target costs more. | Output tokens per second (tokens/s) | Bedrock (LLM inference) | Target scenario | Anthropic Messages API | Difference | | -: | -: | - | -: | -: | | 0 | 300 | Anthropic Messages API | 300 | 0 | | 10 | 694.2 | Anthropic Messages API | 694.2 | 0 | | 100 | 4242 | Anthropic Messages API | 4242 | 0 | | 1000 | 39720 | Anthropic Messages API | 39720 | 0 | | 10000 | 394500 | Anthropic Messages API | 394500 | 0 | | 100000 | 3942300 | Anthropic Messages API | 3942300 | 0 | **Curve assumptions** | Target scenario | Assumption | Value | Unit | | - | - | -: | - | | Anthropic Messages API | Origin list price · Input tokens | 0.000003 | USD/Tokens | | Anthropic Messages API | Origin list price · Output tokens | 0.000015 | USD/Tokens | | Anthropic Messages API | Target / origin input tokens | 1 | multiplier | | Anthropic Messages API | Target / origin output tokens | 1 | multiplier | | Anthropic Messages API | Target price guess · read | 0 | USD/read | | Anthropic Messages API | Target price guess · write | 0 | USD/write | | Anthropic Messages API | Target price guess · provider operation | 0 | USD/provider operation | | Anthropic Messages API | Target price guess · GiB-month | 0 | USD/GiB-month | | Anthropic Messages API | Target price guess · published GiB | 0 | USD/published GiB | | Anthropic Messages API | Target price guess · delivered GiB | 0 | USD/delivered GiB | | Anthropic Messages API | Target price guess · network GiB | 0 | USD/network GiB | | Anthropic Messages API | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | Anthropic Messages API | Target price guess · vCPU-hour | 0 | USD/vCPU-hour | | Anthropic Messages API | Target price guess · GiB-hour | 0 | USD/GiB-hour | | Anthropic Messages API | Target price guess · resource-hour | 0 | USD/resource-hour | | Anthropic Messages API | Target price guess · resource-month | 0 | USD/resource-month | | Anthropic Messages API | Target list price · input token | 0.000003 | USD/input token | | Anthropic Messages API | Target list price · output token | 0.000015 | USD/output token | | Anthropic Messages API | Target monthly base charge | 0 | USD/month | | Anthropic Messages API | Approximate target service minimum | 0 | USD/month | | Anthropic Messages API | Minimum primary CPU | 2 | vCPU | | Anthropic Messages API | Minimum primary RAM | 8 | GiB | | Anthropic Messages API | Minimum primary disk | 100 | GiB | | Anthropic Messages API | Additional full-price standby copy | 0 | 0 or 1 | | Anthropic Messages API | Allocated share of provisioned cost | 1 | fraction | | Anthropic Messages API | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | Anthropic Messages API | Planning peak / mean work | 2 | multiplier | | Anthropic Messages API | Planning primary CPU utilization | 0.7 | fraction | | Anthropic Messages API | Resident state reserved in RAM | 0.1 | fraction | | Anthropic Messages API | Planning disk utilization | 0.8 | fraction | | Anthropic Messages API | Maintenance and log disk reserve | 20 | GiB | | Anthropic Messages API | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | Anthropic Messages API | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | Anthropic Messages API | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | Anthropic Messages API | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | Anthropic Messages API | Dependent object sink write price guess | 0.000005 | USD/write | * Only this native meter varies; all other source usage and translation assumptions retain the displayed defaults. Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. Other native meters retain their explicitly displayed default quantities. * Planning estimate · editable authored rates and sizing assumptions dated 2026-10-22; not verified provider prices. us-east-1 → global, USD/month. * Source prices are reference charges, not an invoice. Target lines cover selected service usage/configuration only; adapter hosting and Tensor9 fees are excluded. * Explicit tariff samples are not performance measurements or capacity qualification. No interpolation or break-even guarantee is implied. * Variant assumptions show the first axis sample; any axis-dependent configuration is described by this curve's workload note. * Rates are cited provider list prices, editable for discounts, commitments or another region. Workload translation and capacity coefficients remain editable planning assumptions, not benchmark measurements or a commercial quote. * Default-zero rates are explicit assumptions, not evidence that the deployed service or resource is free. * Provisioned-unit minimums use an editable approximate allocation floor. Modeled target service cost is max(usage/allocation, minimum), with billable network, durable staging and dependent sinks added separately; actual RU/TU sizing, scaling increments and throughput remain unqualified. * Durable staging is priced once: provisioned targets include adapter state in their selected allocation; native targets add a separate staging allocation when required. Adapter hosting is a separate allocation and must exclude resources already priced as staging, workers or target capacity. * A declared adapter mapping and cost model do not establish implementation availability or service equivalence. * This is a declared directed mapping; a planning estimate does not establish deployment or request-serving support. * Origin unit prices are cited provider list prices; capacity coefficients and translation assumptions are editable authored planning guesses, not measured capacity or a provider quote. * Origin tariffs are the cited Claude Sonnet 4.6 global-endpoint list prices; target tariffs name no asserted current model SKU. Select model-specific rates before using the estimate. * Input and output meters and token expansion remain independent. A cheaper model is not automatically equivalent in capability or quality. * Input and output token tariffs remain distinct and establish no equivalent output quality or tokenizer. Claude Sonnet 4.6, the Claude model the Bedrock adapter's evals and the Vertex Claude arm default to. inference\_geo 'us' applies 1.1x; cache hits (0.30/MTok) and batch (50% off) are not modeled; Claude Sonnet 5 lists 2/10 per MTok. * AWS Price List API, AmazonBedrockFoundationModels offer: [https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json) * Claude API pricing: [https://platform.claude.com/docs/en/about-claude/pricing](https://platform.claude.com/docs/en/about-claude/pricing) Cost scope: Modeled target native service and explicitly modeled state allocation; hosting and Tensor9 fees only when supplied. Excluded costs and workload features: * Prompt caching, batch discounts, context tiers, grounding, tools, agents, embeddings, search, guardrails and dedicated inference hardware. * Taxes, discounts, credits and free-tier allowances unless explicitly reflected in edited rates. * Unmodeled dependent services, task execution and operations outside the selected workload family. List prices read: 2026-10-22. Model: `plan-ccde7cd673441e8ba203e027d2e5f895e072eb7c` / `2026-10-22.planning.2`. Curve: `workload-2`. Sources: [AWS Price List API, AmazonBedrockFoundationModels offer](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json), [Claude API pricing](https://platform.claude.com/docs/en/about-claude/pricing). ## Bedrock model inference → Azure RAI Content Filter Policy Origin API: `aws::1.0.0::bedrock`. Target backend: `azure::1.0.0::rai-policy`. **Native or deployment mapping.** This mapping provisions an Azure content-filter policy. A policy is not a Bedrock inference endpoint; inference adapters have their own operation-specific terms. Cost estimates describe the selected service resources; operation coverage remains separate. ### Bedrock model inference · Input tokens per second Estimated monthly service charges by Input tokens per second. Adapter hosting and Tensor9 fees are excluded. Source: us-east-1; target: Explicit target planning region; currency: USD. ![Bedrock model inference · Input tokens per second: monthly service charges by Input tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/05a6f05cec5ac3a6758249524a9318bc5a4f53fd028ec76d1a7bfb05929311ab-light.svg) ![Bedrock model inference · Input tokens per second: monthly service charges by Input tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/5de4e04e0288759c5741f2e5691033f08b9a756059f939cdc4c151bfa0751540-dark.svg) Comparison pills describe all sampled scale points; card amounts show the named workload. Costs The Same means within 20% at every point. Costs Depend includes mixed results and unpriced samples. Exact sampled model outputs in USD/month. Difference means target minus origin; a positive difference means the target costs more. Estimated allocation from planning assumptions; not benchmark calibrated. Lines connect evaluated samples; intermediate capacity and prices are not inferred. Unpriced means no target quote or cost difference is available within the sizing policy. | Input tokens per second (tokens/s) | Bedrock (LLM inference) | Target scenario | Azure RAI Content Filter Policy | Difference | | -: | -: | - | -: | -: | | 0 | 375 | Azure RAI Content Filter Policy (one copy) | Unpriced | Unavailable | | 0 | 375 | Azure RAI Content Filter Policy (two copies) | Unpriced | Unavailable | | 10 | 453.84 | Azure RAI Content Filter Policy (one copy) | Unpriced | Unavailable | | 10 | 453.84 | Azure RAI Content Filter Policy (two copies) | Unpriced | Unavailable | | 100 | 1163.4 | Azure RAI Content Filter Policy (one copy) | Unpriced | Unavailable | | 100 | 1163.4 | Azure RAI Content Filter Policy (two copies) | Unpriced | Unavailable | | 1000 | 8259 | Azure RAI Content Filter Policy (one copy) | Unpriced | Unavailable | | 1000 | 8259 | Azure RAI Content Filter Policy (two copies) | Unpriced | Unavailable | | 10000 | 79215 | Azure RAI Content Filter Policy (one copy) | Unpriced | Unavailable | | 10000 | 79215 | Azure RAI Content Filter Policy (two copies) | Unpriced | Unavailable | | 100000 | 788775 | Azure RAI Content Filter Policy (one copy) | Unpriced | Unavailable | | 100000 | 788775 | Azure RAI Content Filter Policy (two copies) | Unpriced | Unavailable | **Estimated allocation from planning assumptions; not benchmark calibrated.** The sizing policy grows compute with translated work and peak headroom, memory with compute and working set, and disk with resident working state and reserve. Monthly usage is distinct from resident data; each workload's assumptions define the translation. HA changes pricing, not this policy's throughput estimate. Unpriced workloads exceed the policy's configuration range; no silent vertical cap or automatic sharding is assumed. | Sample / scenario | Basis | Mean / peak work/s | Required vCPU | Allocated vCPU | Memory GiB | SSD GiB | Resident GiB | Policy | Constraints | | - | - | -: | -: | -: | -: | -: | -: | - | - | | 0 tokens/s / Azure RAI Content Filter Policy (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 0 tokens/s / Azure RAI Content Filter Policy (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10 tokens/s / Azure RAI Content Filter Policy (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10 tokens/s / Azure RAI Content Filter Policy (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100 tokens/s / Azure RAI Content Filter Policy (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100 tokens/s / Azure RAI Content Filter Policy (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 1000 tokens/s / Azure RAI Content Filter Policy (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 1000 tokens/s / Azure RAI Content Filter Policy (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10000 tokens/s / Azure RAI Content Filter Policy (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10000 tokens/s / Azure RAI Content Filter Policy (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100000 tokens/s / Azure RAI Content Filter Policy (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100000 tokens/s / Azure RAI Content Filter Policy (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | * Row-store throughput (capacityOpsPerVcpuSecond, default 1500) is measured-by-tensor9 as a lower bound: the DynamoDB adapter on Cloud SQL db-custom-2-7680 (2 vCPU, PostgreSQL 16, us-west1) served 2,974 ops/s of YCSB-A at 16 threads without recorded saturation (src/main/kotlin/t9/prime/equiv/generated/eval/CloudsqlPostgresqlYcsbAT16GoYcsbV1\_0\_0.kt). Published CPU-bound pgbench reaches 3,101 read-write transactions and 20,289 point reads per vCPU-second (AWS z1d.2xlarge, PostgreSQL 14.5), so actual capacity is likely higher. Reads and writes share one coefficient: at the same adapter version, read-only YCSB-C was not faster than 50%-update YCSB-A. * Log-broker throughput (capacityStreamMiBPerVcpuSecond 20 at 0.5 utilization, 4 GiB RAM per vCPU) is provider-documented by Google Managed Service for Apache Kafka sizing ([https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size](https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size)): write-equivalent bytes are replicated writes plus a quarter of reads. Producer batches below 10 KB lower per-CPU throughput; that penalty is not modeled. * Disk utilization 0.8 is provider-documented headroom below Cloud SQL automatic storage increase, which triggers near 91-95% full ([https://docs.cloud.google.com/sql/docs/postgres/instance-settings](https://docs.cloud.google.com/sql/docs/postgres/instance-settings)). The 20 GiB reserve exceeds PostgreSQL's 1 GB default max\_wal\_size; the remainder of that reserve is unmeasured. * CPU tiers, 4 GiB per vCPU, at most 6.5 GiB per vCPU, 96 vCPU and 65,536 GiB disk follow Cloud SQL Enterprise custom machine limits (provider-documented, same page); 624 GiB is 96 × 6.5. Targets billed per node, unit or throughput unit are sized by their backend instead. * No public evidence: peak/mean work (2) and the RAM working-set fraction (0.1). Proposed measurements: the p99 one-minute rate over the monthly mean from customer CloudWatch metrics; the buffer hit ratio while stepping RAM under a fixed YCSB-B load. * Stored GiB already includes the origin family's physical footprint assumptions; no second storage amplification is applied. * The operations/vCPU coefficient supplies the compute requirement alongside explicit resource-hour minima. * Copies and allocation fractions change cost only; they do not increase modeled primary capacity. **Curve assumptions** | Target scenario | Assumption | Value | Unit | | - | - | -: | - | | Azure RAI Content Filter Policy (one copy) | Origin list price · Input tokens | 0.000003 | USD/Tokens | | Azure RAI Content Filter Policy (one copy) | Origin list price · Output tokens | 0.000015 | USD/Tokens | | Azure RAI Content Filter Policy (one copy) | Target / origin input tokens | 1 | multiplier | | Azure RAI Content Filter Policy (one copy) | Target / origin output tokens | 1 | multiplier | | Azure RAI Content Filter Policy (one copy) | Target price guess · read | 0 | USD/read | | Azure RAI Content Filter Policy (one copy) | Target price guess · write | 0 | USD/write | | Azure RAI Content Filter Policy (one copy) | Target price guess · provider operation | 0 | USD/provider operation | | Azure RAI Content Filter Policy (one copy) | Target price guess · GiB-month | 0.17 | USD/GiB-month | | Azure RAI Content Filter Policy (one copy) | Target price guess · published GiB | 0 | USD/published GiB | | Azure RAI Content Filter Policy (one copy) | Target price guess · delivered GiB | 0 | USD/delivered GiB | | Azure RAI Content Filter Policy (one copy) | Target price guess · network GiB | 0.09 | USD/network GiB | | Azure RAI Content Filter Policy (one copy) | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | Azure RAI Content Filter Policy (one copy) | Target price guess · vCPU-hour | 0.05 | USD/vCPU-hour | | Azure RAI Content Filter Policy (one copy) | Target price guess · GiB-hour | 0.006 | USD/GiB-hour | | Azure RAI Content Filter Policy (one copy) | Target price guess · resource-hour | 0 | USD/resource-hour | | Azure RAI Content Filter Policy (one copy) | Target price guess · resource-month | 0 | USD/resource-month | | Azure RAI Content Filter Policy (one copy) | Target price guess · input token | 0 | USD/input token | | Azure RAI Content Filter Policy (one copy) | Target price guess · output token | 0 | USD/output token | | Azure RAI Content Filter Policy (one copy) | Target monthly base charge | 0 | USD/month | | Azure RAI Content Filter Policy (one copy) | Approximate target service minimum | 0 | USD/month | | Azure RAI Content Filter Policy (one copy) | Minimum primary CPU | 2 | vCPU | | Azure RAI Content Filter Policy (one copy) | Minimum primary RAM | 8 | GiB | | Azure RAI Content Filter Policy (one copy) | Minimum primary disk | 100 | GiB | | Azure RAI Content Filter Policy (one copy) | Additional full-price standby copy | 0 | 0 or 1 | | Azure RAI Content Filter Policy (one copy) | Allocated share of provisioned cost | 1 | fraction | | Azure RAI Content Filter Policy (one copy) | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | Azure RAI Content Filter Policy (one copy) | Planning peak / mean work | 2 | multiplier | | Azure RAI Content Filter Policy (one copy) | Planning primary CPU utilization | 0.7 | fraction | | Azure RAI Content Filter Policy (one copy) | Resident state reserved in RAM | 0.1 | fraction | | Azure RAI Content Filter Policy (one copy) | Planning disk utilization | 0.8 | fraction | | Azure RAI Content Filter Policy (one copy) | Maintenance and log disk reserve | 20 | GiB | | Azure RAI Content Filter Policy (one copy) | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | Azure RAI Content Filter Policy (one copy) | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | Azure RAI Content Filter Policy (one copy) | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | Azure RAI Content Filter Policy (one copy) | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | Azure RAI Content Filter Policy (one copy) | Dependent object sink write price guess | 0.000005 | USD/write | | Azure RAI Content Filter Policy (two copies) | Origin list price · Input tokens | 0.000003 | USD/Tokens | | Azure RAI Content Filter Policy (two copies) | Origin list price · Output tokens | 0.000015 | USD/Tokens | | Azure RAI Content Filter Policy (two copies) | Target / origin input tokens | 1 | multiplier | | Azure RAI Content Filter Policy (two copies) | Target / origin output tokens | 1 | multiplier | | Azure RAI Content Filter Policy (two copies) | Target price guess · read | 0 | USD/read | | Azure RAI Content Filter Policy (two copies) | Target price guess · write | 0 | USD/write | | Azure RAI Content Filter Policy (two copies) | Target price guess · provider operation | 0 | USD/provider operation | | Azure RAI Content Filter Policy (two copies) | Target price guess · GiB-month | 0.17 | USD/GiB-month | | Azure RAI Content Filter Policy (two copies) | Target price guess · published GiB | 0 | USD/published GiB | | Azure RAI Content Filter Policy (two copies) | Target price guess · delivered GiB | 0 | USD/delivered GiB | | Azure RAI Content Filter Policy (two copies) | Target price guess · network GiB | 0.09 | USD/network GiB | | Azure RAI Content Filter Policy (two copies) | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | Azure RAI Content Filter Policy (two copies) | Target price guess · vCPU-hour | 0.05 | USD/vCPU-hour | | Azure RAI Content Filter Policy (two copies) | Target price guess · GiB-hour | 0.006 | USD/GiB-hour | | Azure RAI Content Filter Policy (two copies) | Target price guess · resource-hour | 0 | USD/resource-hour | | Azure RAI Content Filter Policy (two copies) | Target price guess · resource-month | 0 | USD/resource-month | | Azure RAI Content Filter Policy (two copies) | Target price guess · input token | 0 | USD/input token | | Azure RAI Content Filter Policy (two copies) | Target price guess · output token | 0 | USD/output token | | Azure RAI Content Filter Policy (two copies) | Target monthly base charge | 0 | USD/month | | Azure RAI Content Filter Policy (two copies) | Approximate target service minimum | 0 | USD/month | | Azure RAI Content Filter Policy (two copies) | Minimum primary CPU | 2 | vCPU | | Azure RAI Content Filter Policy (two copies) | Minimum primary RAM | 8 | GiB | | Azure RAI Content Filter Policy (two copies) | Minimum primary disk | 100 | GiB | | Azure RAI Content Filter Policy (two copies) | Additional full-price standby copy | 1 | 0 or 1 | | Azure RAI Content Filter Policy (two copies) | Allocated share of provisioned cost | 1 | fraction | | Azure RAI Content Filter Policy (two copies) | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | Azure RAI Content Filter Policy (two copies) | Planning peak / mean work | 2 | multiplier | | Azure RAI Content Filter Policy (two copies) | Planning primary CPU utilization | 0.7 | fraction | | Azure RAI Content Filter Policy (two copies) | Resident state reserved in RAM | 0.1 | fraction | | Azure RAI Content Filter Policy (two copies) | Planning disk utilization | 0.8 | fraction | | Azure RAI Content Filter Policy (two copies) | Maintenance and log disk reserve | 20 | GiB | | Azure RAI Content Filter Policy (two copies) | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | Azure RAI Content Filter Policy (two copies) | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | Azure RAI Content Filter Policy (two copies) | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | Azure RAI Content Filter Policy (two copies) | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | Azure RAI Content Filter Policy (two copies) | Dependent object sink write price guess | 0.000005 | USD/write | * Only this native meter varies; all other source usage and translation assumptions retain the displayed defaults. Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. Other native meters retain their explicitly displayed default quantities. * Planning estimate · editable authored rates and sizing assumptions dated 2026-10-22; not verified provider prices. us-east-1 → Explicit target planning region, USD/month. * Source prices are reference charges, not an invoice. Target lines cover selected service usage/configuration only; adapter hosting and Tensor9 fees are excluded. * Explicit tariff samples are not performance measurements or capacity qualification. No interpolation or break-even guarantee is implied. * Variant assumptions show the first axis sample; any axis-dependent configuration is described by this curve's workload note. * Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. * Default-zero rates are explicit assumptions, not evidence that the deployed service or resource is free. * Provisioned-unit minimums use an editable approximate allocation floor. Modeled target service cost is max(usage/allocation, minimum), with billable network, durable staging and dependent sinks added separately; actual RU/TU sizing, scaling increments and throughput remain unqualified. * Durable staging is priced once: provisioned targets include adapter state in their selected allocation; native targets add a separate staging allocation when required. Adapter hosting is a separate allocation and must exclude resources already priced as staging, workers or target capacity. * A declared adapter mapping and cost model do not establish implementation availability or service equivalence. * This is a declared directed mapping; a planning estimate does not establish deployment or request-serving support. * Native/deployment comparison scope: This mapping provisions an Azure content-filter policy. A policy is not a Bedrock inference endpoint; inference adapters have their own operation-specific terms. * Origin unit prices are cited provider list prices; capacity coefficients and translation assumptions are editable authored planning guesses, not measured capacity or a provider quote. * Origin tariffs are the cited Claude Sonnet 4.6 global-endpoint list prices; target tariffs name no asserted current model SKU. Select model-specific rates before using the estimate. * Input and output meters and token expansion remain independent. A cheaper model is not automatically equivalent in capability or quality. * All target rates are editable authored planning guesses, not verified current list prices, customer discounts or benchmark evidence. * Dedicated accelerator/model-fit target: the Bedrock family returns this direction as Unpriced because the serving GPU or model deployment is not selected, so these generic rates are not shown as list prices. Scaleway Managed Inference lists EUR 0.93/hour for an L4-1-24G node ([https://www.scaleway.com/en/pricing/model-as-a-service/](https://www.scaleway.com/en/pricing/model-as-a-service/)); the Azure RAI policy has no meter of its own, its tokens bill on the Azure OpenAI deployment and knowledge bases on Azure AI Search (Basic USD 0.101/hour); the Tensor9 in-cluster model service needs GPU node pricing. Read 2026-10-22. * Row-store throughput (capacityOpsPerVcpuSecond, default 1500) is measured-by-tensor9 as a lower bound: the DynamoDB adapter on Cloud SQL db-custom-2-7680 (2 vCPU, PostgreSQL 16, us-west1) served 2,974 ops/s of YCSB-A at 16 threads without recorded saturation (src/main/kotlin/t9/prime/equiv/generated/eval/CloudsqlPostgresqlYcsbAT16GoYcsbV1\_0\_0.kt). Published CPU-bound pgbench reaches 3,101 read-write transactions and 20,289 point reads per vCPU-second (AWS z1d.2xlarge, PostgreSQL 14.5), so actual capacity is likely higher. Reads and writes share one coefficient: at the same adapter version, read-only YCSB-C was not faster than 50%-update YCSB-A. * Log-broker throughput (capacityStreamMiBPerVcpuSecond 20 at 0.5 utilization, 4 GiB RAM per vCPU) is provider-documented by Google Managed Service for Apache Kafka sizing ([https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size](https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size)): write-equivalent bytes are replicated writes plus a quarter of reads. Producer batches below 10 KB lower per-CPU throughput; that penalty is not modeled. * Disk utilization 0.8 is provider-documented headroom below Cloud SQL automatic storage increase, which triggers near 91-95% full ([https://docs.cloud.google.com/sql/docs/postgres/instance-settings](https://docs.cloud.google.com/sql/docs/postgres/instance-settings)). The 20 GiB reserve exceeds PostgreSQL's 1 GB default max\_wal\_size; the remainder of that reserve is unmeasured. * CPU tiers, 4 GiB per vCPU, at most 6.5 GiB per vCPU, 96 vCPU and 65,536 GiB disk follow Cloud SQL Enterprise custom machine limits (provider-documented, same page); 624 GiB is 96 × 6.5. Targets billed per node, unit or throughput unit are sized by their backend instead. * No public evidence: peak/mean work (2) and the RAM working-set fraction (0.1). Proposed measurements: the p99 one-minute rate over the monthly mean from customer CloudWatch metrics; the buffer hit ratio while stepping RAM under a fixed YCSB-B load. * Stored GiB already includes the origin family's physical footprint assumptions; no second storage amplification is applied. * The operations/vCPU coefficient supplies the compute requirement alongside explicit resource-hour minima. * Copies and allocation fractions change cost only; they do not increase modeled primary capacity. * AWS Price List API, AmazonBedrockFoundationModels offer: [https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json) Cost scope: Modeled target native service and explicitly modeled state allocation; hosting and Tensor9 fees only when supplied. Excluded costs and workload features: * Prompt caching, batch discounts, context tiers, grounding, tools, agents, embeddings, search, guardrails and dedicated inference hardware. * Taxes, discounts, credits and free-tier allowances unless explicitly reflected in edited rates. * Unmodeled dependent services, task execution and operations outside the selected workload family. List prices read: 2026-10-22. Model: `plan-16611989e2e34abf6767c0c4525fdf4ae72b8f64` / `2026-10-22.planning.2`. Curve: `workload-1`. Sources: [AWS Price List API, AmazonBedrockFoundationModels offer](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json). ### Bedrock model inference · Output tokens per second Estimated monthly service charges by Output tokens per second. Adapter hosting and Tensor9 fees are excluded. Source: us-east-1; target: Explicit target planning region; currency: USD. ![Bedrock model inference · Output tokens per second: monthly service charges by Output tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/f8509f2bb5430682d1ff53ab4a8a765b706c4d2702213345cb1ab99e5faff225-light.svg) ![Bedrock model inference · Output tokens per second: monthly service charges by Output tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/8f47501a18ae3e32da51659171eaf3df118d4144c358c6ac00bed776661e336e-dark.svg) Comparison pills describe all sampled scale points; card amounts show the named workload. Costs The Same means within 20% at every point. Costs Depend includes mixed results and unpriced samples. Exact sampled model outputs in USD/month. Difference means target minus origin; a positive difference means the target costs more. Estimated allocation from planning assumptions; not benchmark calibrated. Lines connect evaluated samples; intermediate capacity and prices are not inferred. Unpriced means no target quote or cost difference is available within the sizing policy. | Output tokens per second (tokens/s) | Bedrock (LLM inference) | Target scenario | Azure RAI Content Filter Policy | Difference | | -: | -: | - | -: | -: | | 0 | 300 | Azure RAI Content Filter Policy (one copy) | Unpriced | Unavailable | | 0 | 300 | Azure RAI Content Filter Policy (two copies) | Unpriced | Unavailable | | 10 | 694.2 | Azure RAI Content Filter Policy (one copy) | Unpriced | Unavailable | | 10 | 694.2 | Azure RAI Content Filter Policy (two copies) | Unpriced | Unavailable | | 100 | 4242 | Azure RAI Content Filter Policy (one copy) | Unpriced | Unavailable | | 100 | 4242 | Azure RAI Content Filter Policy (two copies) | Unpriced | Unavailable | | 1000 | 39720 | Azure RAI Content Filter Policy (one copy) | Unpriced | Unavailable | | 1000 | 39720 | Azure RAI Content Filter Policy (two copies) | Unpriced | Unavailable | | 10000 | 394500 | Azure RAI Content Filter Policy (one copy) | Unpriced | Unavailable | | 10000 | 394500 | Azure RAI Content Filter Policy (two copies) | Unpriced | Unavailable | | 100000 | 3942300 | Azure RAI Content Filter Policy (one copy) | Unpriced | Unavailable | | 100000 | 3942300 | Azure RAI Content Filter Policy (two copies) | Unpriced | Unavailable | **Estimated allocation from planning assumptions; not benchmark calibrated.** The sizing policy grows compute with translated work and peak headroom, memory with compute and working set, and disk with resident working state and reserve. Monthly usage is distinct from resident data; each workload's assumptions define the translation. HA changes pricing, not this policy's throughput estimate. Unpriced workloads exceed the policy's configuration range; no silent vertical cap or automatic sharding is assumed. | Sample / scenario | Basis | Mean / peak work/s | Required vCPU | Allocated vCPU | Memory GiB | SSD GiB | Resident GiB | Policy | Constraints | | - | - | -: | -: | -: | -: | -: | -: | - | - | | 0 tokens/s / Azure RAI Content Filter Policy (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 0 tokens/s / Azure RAI Content Filter Policy (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10 tokens/s / Azure RAI Content Filter Policy (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10 tokens/s / Azure RAI Content Filter Policy (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100 tokens/s / Azure RAI Content Filter Policy (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100 tokens/s / Azure RAI Content Filter Policy (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 1000 tokens/s / Azure RAI Content Filter Policy (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 1000 tokens/s / Azure RAI Content Filter Policy (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10000 tokens/s / Azure RAI Content Filter Policy (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10000 tokens/s / Azure RAI Content Filter Policy (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100000 tokens/s / Azure RAI Content Filter Policy (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100000 tokens/s / Azure RAI Content Filter Policy (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | * Row-store throughput (capacityOpsPerVcpuSecond, default 1500) is measured-by-tensor9 as a lower bound: the DynamoDB adapter on Cloud SQL db-custom-2-7680 (2 vCPU, PostgreSQL 16, us-west1) served 2,974 ops/s of YCSB-A at 16 threads without recorded saturation (src/main/kotlin/t9/prime/equiv/generated/eval/CloudsqlPostgresqlYcsbAT16GoYcsbV1\_0\_0.kt). Published CPU-bound pgbench reaches 3,101 read-write transactions and 20,289 point reads per vCPU-second (AWS z1d.2xlarge, PostgreSQL 14.5), so actual capacity is likely higher. Reads and writes share one coefficient: at the same adapter version, read-only YCSB-C was not faster than 50%-update YCSB-A. * Log-broker throughput (capacityStreamMiBPerVcpuSecond 20 at 0.5 utilization, 4 GiB RAM per vCPU) is provider-documented by Google Managed Service for Apache Kafka sizing ([https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size](https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size)): write-equivalent bytes are replicated writes plus a quarter of reads. Producer batches below 10 KB lower per-CPU throughput; that penalty is not modeled. * Disk utilization 0.8 is provider-documented headroom below Cloud SQL automatic storage increase, which triggers near 91-95% full ([https://docs.cloud.google.com/sql/docs/postgres/instance-settings](https://docs.cloud.google.com/sql/docs/postgres/instance-settings)). The 20 GiB reserve exceeds PostgreSQL's 1 GB default max\_wal\_size; the remainder of that reserve is unmeasured. * CPU tiers, 4 GiB per vCPU, at most 6.5 GiB per vCPU, 96 vCPU and 65,536 GiB disk follow Cloud SQL Enterprise custom machine limits (provider-documented, same page); 624 GiB is 96 × 6.5. Targets billed per node, unit or throughput unit are sized by their backend instead. * No public evidence: peak/mean work (2) and the RAM working-set fraction (0.1). Proposed measurements: the p99 one-minute rate over the monthly mean from customer CloudWatch metrics; the buffer hit ratio while stepping RAM under a fixed YCSB-B load. * Stored GiB already includes the origin family's physical footprint assumptions; no second storage amplification is applied. * The operations/vCPU coefficient supplies the compute requirement alongside explicit resource-hour minima. * Copies and allocation fractions change cost only; they do not increase modeled primary capacity. **Curve assumptions** | Target scenario | Assumption | Value | Unit | | - | - | -: | - | | Azure RAI Content Filter Policy (one copy) | Origin list price · Input tokens | 0.000003 | USD/Tokens | | Azure RAI Content Filter Policy (one copy) | Origin list price · Output tokens | 0.000015 | USD/Tokens | | Azure RAI Content Filter Policy (one copy) | Target / origin input tokens | 1 | multiplier | | Azure RAI Content Filter Policy (one copy) | Target / origin output tokens | 1 | multiplier | | Azure RAI Content Filter Policy (one copy) | Target price guess · read | 0 | USD/read | | Azure RAI Content Filter Policy (one copy) | Target price guess · write | 0 | USD/write | | Azure RAI Content Filter Policy (one copy) | Target price guess · provider operation | 0 | USD/provider operation | | Azure RAI Content Filter Policy (one copy) | Target price guess · GiB-month | 0.17 | USD/GiB-month | | Azure RAI Content Filter Policy (one copy) | Target price guess · published GiB | 0 | USD/published GiB | | Azure RAI Content Filter Policy (one copy) | Target price guess · delivered GiB | 0 | USD/delivered GiB | | Azure RAI Content Filter Policy (one copy) | Target price guess · network GiB | 0.09 | USD/network GiB | | Azure RAI Content Filter Policy (one copy) | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | Azure RAI Content Filter Policy (one copy) | Target price guess · vCPU-hour | 0.05 | USD/vCPU-hour | | Azure RAI Content Filter Policy (one copy) | Target price guess · GiB-hour | 0.006 | USD/GiB-hour | | Azure RAI Content Filter Policy (one copy) | Target price guess · resource-hour | 0 | USD/resource-hour | | Azure RAI Content Filter Policy (one copy) | Target price guess · resource-month | 0 | USD/resource-month | | Azure RAI Content Filter Policy (one copy) | Target price guess · input token | 0 | USD/input token | | Azure RAI Content Filter Policy (one copy) | Target price guess · output token | 0 | USD/output token | | Azure RAI Content Filter Policy (one copy) | Target monthly base charge | 0 | USD/month | | Azure RAI Content Filter Policy (one copy) | Approximate target service minimum | 0 | USD/month | | Azure RAI Content Filter Policy (one copy) | Minimum primary CPU | 2 | vCPU | | Azure RAI Content Filter Policy (one copy) | Minimum primary RAM | 8 | GiB | | Azure RAI Content Filter Policy (one copy) | Minimum primary disk | 100 | GiB | | Azure RAI Content Filter Policy (one copy) | Additional full-price standby copy | 0 | 0 or 1 | | Azure RAI Content Filter Policy (one copy) | Allocated share of provisioned cost | 1 | fraction | | Azure RAI Content Filter Policy (one copy) | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | Azure RAI Content Filter Policy (one copy) | Planning peak / mean work | 2 | multiplier | | Azure RAI Content Filter Policy (one copy) | Planning primary CPU utilization | 0.7 | fraction | | Azure RAI Content Filter Policy (one copy) | Resident state reserved in RAM | 0.1 | fraction | | Azure RAI Content Filter Policy (one copy) | Planning disk utilization | 0.8 | fraction | | Azure RAI Content Filter Policy (one copy) | Maintenance and log disk reserve | 20 | GiB | | Azure RAI Content Filter Policy (one copy) | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | Azure RAI Content Filter Policy (one copy) | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | Azure RAI Content Filter Policy (one copy) | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | Azure RAI Content Filter Policy (one copy) | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | Azure RAI Content Filter Policy (one copy) | Dependent object sink write price guess | 0.000005 | USD/write | | Azure RAI Content Filter Policy (two copies) | Origin list price · Input tokens | 0.000003 | USD/Tokens | | Azure RAI Content Filter Policy (two copies) | Origin list price · Output tokens | 0.000015 | USD/Tokens | | Azure RAI Content Filter Policy (two copies) | Target / origin input tokens | 1 | multiplier | | Azure RAI Content Filter Policy (two copies) | Target / origin output tokens | 1 | multiplier | | Azure RAI Content Filter Policy (two copies) | Target price guess · read | 0 | USD/read | | Azure RAI Content Filter Policy (two copies) | Target price guess · write | 0 | USD/write | | Azure RAI Content Filter Policy (two copies) | Target price guess · provider operation | 0 | USD/provider operation | | Azure RAI Content Filter Policy (two copies) | Target price guess · GiB-month | 0.17 | USD/GiB-month | | Azure RAI Content Filter Policy (two copies) | Target price guess · published GiB | 0 | USD/published GiB | | Azure RAI Content Filter Policy (two copies) | Target price guess · delivered GiB | 0 | USD/delivered GiB | | Azure RAI Content Filter Policy (two copies) | Target price guess · network GiB | 0.09 | USD/network GiB | | Azure RAI Content Filter Policy (two copies) | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | Azure RAI Content Filter Policy (two copies) | Target price guess · vCPU-hour | 0.05 | USD/vCPU-hour | | Azure RAI Content Filter Policy (two copies) | Target price guess · GiB-hour | 0.006 | USD/GiB-hour | | Azure RAI Content Filter Policy (two copies) | Target price guess · resource-hour | 0 | USD/resource-hour | | Azure RAI Content Filter Policy (two copies) | Target price guess · resource-month | 0 | USD/resource-month | | Azure RAI Content Filter Policy (two copies) | Target price guess · input token | 0 | USD/input token | | Azure RAI Content Filter Policy (two copies) | Target price guess · output token | 0 | USD/output token | | Azure RAI Content Filter Policy (two copies) | Target monthly base charge | 0 | USD/month | | Azure RAI Content Filter Policy (two copies) | Approximate target service minimum | 0 | USD/month | | Azure RAI Content Filter Policy (two copies) | Minimum primary CPU | 2 | vCPU | | Azure RAI Content Filter Policy (two copies) | Minimum primary RAM | 8 | GiB | | Azure RAI Content Filter Policy (two copies) | Minimum primary disk | 100 | GiB | | Azure RAI Content Filter Policy (two copies) | Additional full-price standby copy | 1 | 0 or 1 | | Azure RAI Content Filter Policy (two copies) | Allocated share of provisioned cost | 1 | fraction | | Azure RAI Content Filter Policy (two copies) | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | Azure RAI Content Filter Policy (two copies) | Planning peak / mean work | 2 | multiplier | | Azure RAI Content Filter Policy (two copies) | Planning primary CPU utilization | 0.7 | fraction | | Azure RAI Content Filter Policy (two copies) | Resident state reserved in RAM | 0.1 | fraction | | Azure RAI Content Filter Policy (two copies) | Planning disk utilization | 0.8 | fraction | | Azure RAI Content Filter Policy (two copies) | Maintenance and log disk reserve | 20 | GiB | | Azure RAI Content Filter Policy (two copies) | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | Azure RAI Content Filter Policy (two copies) | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | Azure RAI Content Filter Policy (two copies) | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | Azure RAI Content Filter Policy (two copies) | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | Azure RAI Content Filter Policy (two copies) | Dependent object sink write price guess | 0.000005 | USD/write | * Only this native meter varies; all other source usage and translation assumptions retain the displayed defaults. Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. Other native meters retain their explicitly displayed default quantities. * Planning estimate · editable authored rates and sizing assumptions dated 2026-10-22; not verified provider prices. us-east-1 → Explicit target planning region, USD/month. * Source prices are reference charges, not an invoice. Target lines cover selected service usage/configuration only; adapter hosting and Tensor9 fees are excluded. * Explicit tariff samples are not performance measurements or capacity qualification. No interpolation or break-even guarantee is implied. * Variant assumptions show the first axis sample; any axis-dependent configuration is described by this curve's workload note. * Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. * Default-zero rates are explicit assumptions, not evidence that the deployed service or resource is free. * Provisioned-unit minimums use an editable approximate allocation floor. Modeled target service cost is max(usage/allocation, minimum), with billable network, durable staging and dependent sinks added separately; actual RU/TU sizing, scaling increments and throughput remain unqualified. * Durable staging is priced once: provisioned targets include adapter state in their selected allocation; native targets add a separate staging allocation when required. Adapter hosting is a separate allocation and must exclude resources already priced as staging, workers or target capacity. * A declared adapter mapping and cost model do not establish implementation availability or service equivalence. * This is a declared directed mapping; a planning estimate does not establish deployment or request-serving support. * Native/deployment comparison scope: This mapping provisions an Azure content-filter policy. A policy is not a Bedrock inference endpoint; inference adapters have their own operation-specific terms. * Origin unit prices are cited provider list prices; capacity coefficients and translation assumptions are editable authored planning guesses, not measured capacity or a provider quote. * Origin tariffs are the cited Claude Sonnet 4.6 global-endpoint list prices; target tariffs name no asserted current model SKU. Select model-specific rates before using the estimate. * Input and output meters and token expansion remain independent. A cheaper model is not automatically equivalent in capability or quality. * All target rates are editable authored planning guesses, not verified current list prices, customer discounts or benchmark evidence. * Dedicated accelerator/model-fit target: the Bedrock family returns this direction as Unpriced because the serving GPU or model deployment is not selected, so these generic rates are not shown as list prices. Scaleway Managed Inference lists EUR 0.93/hour for an L4-1-24G node ([https://www.scaleway.com/en/pricing/model-as-a-service/](https://www.scaleway.com/en/pricing/model-as-a-service/)); the Azure RAI policy has no meter of its own, its tokens bill on the Azure OpenAI deployment and knowledge bases on Azure AI Search (Basic USD 0.101/hour); the Tensor9 in-cluster model service needs GPU node pricing. Read 2026-10-22. * Row-store throughput (capacityOpsPerVcpuSecond, default 1500) is measured-by-tensor9 as a lower bound: the DynamoDB adapter on Cloud SQL db-custom-2-7680 (2 vCPU, PostgreSQL 16, us-west1) served 2,974 ops/s of YCSB-A at 16 threads without recorded saturation (src/main/kotlin/t9/prime/equiv/generated/eval/CloudsqlPostgresqlYcsbAT16GoYcsbV1\_0\_0.kt). Published CPU-bound pgbench reaches 3,101 read-write transactions and 20,289 point reads per vCPU-second (AWS z1d.2xlarge, PostgreSQL 14.5), so actual capacity is likely higher. Reads and writes share one coefficient: at the same adapter version, read-only YCSB-C was not faster than 50%-update YCSB-A. * Log-broker throughput (capacityStreamMiBPerVcpuSecond 20 at 0.5 utilization, 4 GiB RAM per vCPU) is provider-documented by Google Managed Service for Apache Kafka sizing ([https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size](https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size)): write-equivalent bytes are replicated writes plus a quarter of reads. Producer batches below 10 KB lower per-CPU throughput; that penalty is not modeled. * Disk utilization 0.8 is provider-documented headroom below Cloud SQL automatic storage increase, which triggers near 91-95% full ([https://docs.cloud.google.com/sql/docs/postgres/instance-settings](https://docs.cloud.google.com/sql/docs/postgres/instance-settings)). The 20 GiB reserve exceeds PostgreSQL's 1 GB default max\_wal\_size; the remainder of that reserve is unmeasured. * CPU tiers, 4 GiB per vCPU, at most 6.5 GiB per vCPU, 96 vCPU and 65,536 GiB disk follow Cloud SQL Enterprise custom machine limits (provider-documented, same page); 624 GiB is 96 × 6.5. Targets billed per node, unit or throughput unit are sized by their backend instead. * No public evidence: peak/mean work (2) and the RAM working-set fraction (0.1). Proposed measurements: the p99 one-minute rate over the monthly mean from customer CloudWatch metrics; the buffer hit ratio while stepping RAM under a fixed YCSB-B load. * Stored GiB already includes the origin family's physical footprint assumptions; no second storage amplification is applied. * The operations/vCPU coefficient supplies the compute requirement alongside explicit resource-hour minima. * Copies and allocation fractions change cost only; they do not increase modeled primary capacity. * AWS Price List API, AmazonBedrockFoundationModels offer: [https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json) Cost scope: Modeled target native service and explicitly modeled state allocation; hosting and Tensor9 fees only when supplied. Excluded costs and workload features: * Prompt caching, batch discounts, context tiers, grounding, tools, agents, embeddings, search, guardrails and dedicated inference hardware. * Taxes, discounts, credits and free-tier allowances unless explicitly reflected in edited rates. * Unmodeled dependent services, task execution and operations outside the selected workload family. List prices read: 2026-10-22. Model: `plan-16611989e2e34abf6767c0c4525fdf4ae72b8f64` / `2026-10-22.planning.2`. Curve: `workload-2`. Sources: [AWS Price List API, AmazonBedrockFoundationModels offer](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json). ## Bedrock model inference → OCI Generative AI Origin API: `aws::1.0.0::bedrock`. Target backend: `oci::1.0.0::genai`. ### Bedrock model inference · Input tokens per second Estimated monthly service charges by Input tokens per second. Adapter hosting and Tensor9 fees are excluded. Source: us-east-1; target: us-ashburn-1; currency: USD. ![Bedrock model inference · Input tokens per second: monthly service charges by Input tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/a3e6a8391a1ceb6bdc64fd963cfca5ffdaf88543914559243ed979de77ad7619-light.svg) ![Bedrock model inference · Input tokens per second: monthly service charges by Input tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/347c2f02c95a15754e17236e4002a1fc4384f1082b2945be12f8726276e9d84d-dark.svg) Comparison pills describe all sampled scale points; card amounts show the named workload. Costs The Same means within 20% at every point. Costs Depend includes mixed results and unpriced samples. Exact sampled model outputs in USD/month. Difference means target minus origin; a positive difference means the target costs more. | Input tokens per second (tokens/s) | Bedrock (LLM inference) | Target scenario | OCI Generative AI | Difference | | -: | -: | - | -: | -: | | 0 | 375 | OCI Generative AI | 18 | -357 | | 10 | 453.84 | OCI Generative AI | 36.9216 | -416.9184 | | 100 | 1163.4 | OCI Generative AI | 207.216 | -956.184 | | 1000 | 8259 | OCI Generative AI | 1910.16 | -6348.84 | | 10000 | 79215 | OCI Generative AI | 18939.6 | -60275.4 | | 100000 | 788775 | OCI Generative AI | 189234 | -599541 | **Curve assumptions** | Target scenario | Assumption | Value | Unit | | - | - | -: | - | | OCI Generative AI | Origin list price · Input tokens | 0.000003 | USD/Tokens | | OCI Generative AI | Origin list price · Output tokens | 0.000015 | USD/Tokens | | OCI Generative AI | Target / origin input tokens | 1 | multiplier | | OCI Generative AI | Target / origin output tokens | 1 | multiplier | | OCI Generative AI | Target price guess · read | 0 | USD/read | | OCI Generative AI | Target price guess · write | 0 | USD/write | | OCI Generative AI | Target price guess · provider operation | 0 | USD/provider operation | | OCI Generative AI | Target price guess · GiB-month | 0 | USD/GiB-month | | OCI Generative AI | Target price guess · published GiB | 0 | USD/published GiB | | OCI Generative AI | Target price guess · delivered GiB | 0 | USD/delivered GiB | | OCI Generative AI | Target price guess · network GiB | 0 | USD/network GiB | | OCI Generative AI | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | OCI Generative AI | Target price guess · vCPU-hour | 0 | USD/vCPU-hour | | OCI Generative AI | Target price guess · GiB-hour | 0 | USD/GiB-hour | | OCI Generative AI | Target price guess · resource-hour | 0 | USD/resource-hour | | OCI Generative AI | Target price guess · resource-month | 0 | USD/resource-month | | OCI Generative AI | Target list price · input token | 0.00000072 | USD/input token | | OCI Generative AI | Target list price · output token | 0.00000072 | USD/output token | | OCI Generative AI | Target monthly base charge | 0 | USD/month | | OCI Generative AI | Approximate target service minimum | 0 | USD/month | | OCI Generative AI | Minimum primary CPU | 2 | vCPU | | OCI Generative AI | Minimum primary RAM | 8 | GiB | | OCI Generative AI | Minimum primary disk | 100 | GiB | | OCI Generative AI | Additional full-price standby copy | 0 | 0 or 1 | | OCI Generative AI | Allocated share of provisioned cost | 1 | fraction | | OCI Generative AI | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | OCI Generative AI | Planning peak / mean work | 2 | multiplier | | OCI Generative AI | Planning primary CPU utilization | 0.7 | fraction | | OCI Generative AI | Resident state reserved in RAM | 0.1 | fraction | | OCI Generative AI | Planning disk utilization | 0.8 | fraction | | OCI Generative AI | Maintenance and log disk reserve | 20 | GiB | | OCI Generative AI | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | OCI Generative AI | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | OCI Generative AI | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | OCI Generative AI | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | OCI Generative AI | Dependent object sink write price guess | 0.000005 | USD/write | * Only this native meter varies; all other source usage and translation assumptions retain the displayed defaults. Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. Other native meters retain their explicitly displayed default quantities. * Planning estimate · editable authored rates and sizing assumptions dated 2026-10-22; not verified provider prices. us-east-1 → us-ashburn-1, USD/month. * Source prices are reference charges, not an invoice. Target lines cover selected service usage/configuration only; adapter hosting and Tensor9 fees are excluded. * Explicit tariff samples are not performance measurements or capacity qualification. No interpolation or break-even guarantee is implied. * Variant assumptions show the first axis sample; any axis-dependent configuration is described by this curve's workload note. * Rates are cited provider list prices, editable for discounts, commitments or another region. Workload translation and capacity coefficients remain editable planning assumptions, not benchmark measurements or a commercial quote. * Default-zero rates are explicit assumptions, not evidence that the deployed service or resource is free. * Provisioned-unit minimums use an editable approximate allocation floor. Modeled target service cost is max(usage/allocation, minimum), with billable network, durable staging and dependent sinks added separately; actual RU/TU sizing, scaling increments and throughput remain unqualified. * Durable staging is priced once: provisioned targets include adapter state in their selected allocation; native targets add a separate staging allocation when required. Adapter hosting is a separate allocation and must exclude resources already priced as staging, workers or target capacity. * A declared adapter mapping and cost model do not establish implementation availability or service equivalence. * This is a declared directed mapping; a planning estimate does not establish deployment or request-serving support. * Origin unit prices are cited provider list prices; capacity coefficients and translation assumptions are editable authored planning guesses, not measured capacity or a provider quote. * Origin tariffs are the cited Claude Sonnet 4.6 global-endpoint list prices; target tariffs name no asserted current model SKU. Select model-specific rates before using the estimate. * Input and output meters and token expansion remain independent. A cheaper model is not automatically equivalent in capability or quality. * Input and output token tariffs remain distinct and establish no equivalent output quality or tokenizer. meta.llama-3.3-70b-instruct on demand bills per character of prompt plus response (one transaction is one character, [https://docs.oracle.com/en-us/iaas/Content/generative-ai/pay-on-demand.htm](https://docs.oracle.com/en-us/iaas/Content/generative-ai/pay-on-demand.htm)). The token rates assume 4 characters per token, a planning assumption rather than a listed figure. * AWS Price List API, AmazonBedrockFoundationModels offer: [https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json) * OCI price list API: [https://apexapps.oracle.com/pls/apex/cetools/api/v1/products/](https://apexapps.oracle.com/pls/apex/cetools/api/v1/products/) Cost scope: Modeled target native service and explicitly modeled state allocation; hosting and Tensor9 fees only when supplied. Excluded costs and workload features: * Prompt caching, batch discounts, context tiers, grounding, tools, agents, embeddings, search, guardrails and dedicated inference hardware. * Taxes, discounts, credits and free-tier allowances unless explicitly reflected in edited rates. * Unmodeled dependent services, task execution and operations outside the selected workload family. List prices read: 2026-10-22. Model: `plan-867e301a8e1fa800d4506aa97b8132fa27fdb26a` / `2026-10-22.planning.2`. Curve: `workload-1`. Sources: [AWS Price List API, AmazonBedrockFoundationModels offer](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json), [OCI price list API](https://apexapps.oracle.com/pls/apex/cetools/api/v1/products/). ### Bedrock model inference · Output tokens per second Estimated monthly service charges by Output tokens per second. Adapter hosting and Tensor9 fees are excluded. Source: us-east-1; target: us-ashburn-1; currency: USD. ![Bedrock model inference · Output tokens per second: monthly service charges by Output tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/16478a70243be3d3ddce8f34b64b66c22be0cf3c3814f603135c52648239e7b9-light.svg) ![Bedrock model inference · Output tokens per second: monthly service charges by Output tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/cca7ff6cf1c12dfa17b948ad8cd5830416612077ed3c0d1d4e9878601d45c6dc-dark.svg) Comparison pills describe all sampled scale points; card amounts show the named workload. Costs The Same means within 20% at every point. Costs Depend includes mixed results and unpriced samples. Exact sampled model outputs in USD/month. Difference means target minus origin; a positive difference means the target costs more. | Output tokens per second (tokens/s) | Bedrock (LLM inference) | Target scenario | OCI Generative AI | Difference | | -: | -: | - | -: | -: | | 0 | 300 | OCI Generative AI | 72 | -228 | | 10 | 694.2 | OCI Generative AI | 90.9216 | -603.2784 | | 100 | 4242 | OCI Generative AI | 261.216 | -3980.784 | | 1000 | 39720 | OCI Generative AI | 1964.16 | -37755.84 | | 10000 | 394500 | OCI Generative AI | 18993.6 | -375506.4 | | 100000 | 3942300 | OCI Generative AI | 189288 | -3753012 | **Curve assumptions** | Target scenario | Assumption | Value | Unit | | - | - | -: | - | | OCI Generative AI | Origin list price · Input tokens | 0.000003 | USD/Tokens | | OCI Generative AI | Origin list price · Output tokens | 0.000015 | USD/Tokens | | OCI Generative AI | Target / origin input tokens | 1 | multiplier | | OCI Generative AI | Target / origin output tokens | 1 | multiplier | | OCI Generative AI | Target price guess · read | 0 | USD/read | | OCI Generative AI | Target price guess · write | 0 | USD/write | | OCI Generative AI | Target price guess · provider operation | 0 | USD/provider operation | | OCI Generative AI | Target price guess · GiB-month | 0 | USD/GiB-month | | OCI Generative AI | Target price guess · published GiB | 0 | USD/published GiB | | OCI Generative AI | Target price guess · delivered GiB | 0 | USD/delivered GiB | | OCI Generative AI | Target price guess · network GiB | 0 | USD/network GiB | | OCI Generative AI | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | OCI Generative AI | Target price guess · vCPU-hour | 0 | USD/vCPU-hour | | OCI Generative AI | Target price guess · GiB-hour | 0 | USD/GiB-hour | | OCI Generative AI | Target price guess · resource-hour | 0 | USD/resource-hour | | OCI Generative AI | Target price guess · resource-month | 0 | USD/resource-month | | OCI Generative AI | Target list price · input token | 0.00000072 | USD/input token | | OCI Generative AI | Target list price · output token | 0.00000072 | USD/output token | | OCI Generative AI | Target monthly base charge | 0 | USD/month | | OCI Generative AI | Approximate target service minimum | 0 | USD/month | | OCI Generative AI | Minimum primary CPU | 2 | vCPU | | OCI Generative AI | Minimum primary RAM | 8 | GiB | | OCI Generative AI | Minimum primary disk | 100 | GiB | | OCI Generative AI | Additional full-price standby copy | 0 | 0 or 1 | | OCI Generative AI | Allocated share of provisioned cost | 1 | fraction | | OCI Generative AI | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | OCI Generative AI | Planning peak / mean work | 2 | multiplier | | OCI Generative AI | Planning primary CPU utilization | 0.7 | fraction | | OCI Generative AI | Resident state reserved in RAM | 0.1 | fraction | | OCI Generative AI | Planning disk utilization | 0.8 | fraction | | OCI Generative AI | Maintenance and log disk reserve | 20 | GiB | | OCI Generative AI | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | OCI Generative AI | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | OCI Generative AI | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | OCI Generative AI | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | OCI Generative AI | Dependent object sink write price guess | 0.000005 | USD/write | * Only this native meter varies; all other source usage and translation assumptions retain the displayed defaults. Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. Other native meters retain their explicitly displayed default quantities. * Planning estimate · editable authored rates and sizing assumptions dated 2026-10-22; not verified provider prices. us-east-1 → us-ashburn-1, USD/month. * Source prices are reference charges, not an invoice. Target lines cover selected service usage/configuration only; adapter hosting and Tensor9 fees are excluded. * Explicit tariff samples are not performance measurements or capacity qualification. No interpolation or break-even guarantee is implied. * Variant assumptions show the first axis sample; any axis-dependent configuration is described by this curve's workload note. * Rates are cited provider list prices, editable for discounts, commitments or another region. Workload translation and capacity coefficients remain editable planning assumptions, not benchmark measurements or a commercial quote. * Default-zero rates are explicit assumptions, not evidence that the deployed service or resource is free. * Provisioned-unit minimums use an editable approximate allocation floor. Modeled target service cost is max(usage/allocation, minimum), with billable network, durable staging and dependent sinks added separately; actual RU/TU sizing, scaling increments and throughput remain unqualified. * Durable staging is priced once: provisioned targets include adapter state in their selected allocation; native targets add a separate staging allocation when required. Adapter hosting is a separate allocation and must exclude resources already priced as staging, workers or target capacity. * A declared adapter mapping and cost model do not establish implementation availability or service equivalence. * This is a declared directed mapping; a planning estimate does not establish deployment or request-serving support. * Origin unit prices are cited provider list prices; capacity coefficients and translation assumptions are editable authored planning guesses, not measured capacity or a provider quote. * Origin tariffs are the cited Claude Sonnet 4.6 global-endpoint list prices; target tariffs name no asserted current model SKU. Select model-specific rates before using the estimate. * Input and output meters and token expansion remain independent. A cheaper model is not automatically equivalent in capability or quality. * Input and output token tariffs remain distinct and establish no equivalent output quality or tokenizer. meta.llama-3.3-70b-instruct on demand bills per character of prompt plus response (one transaction is one character, [https://docs.oracle.com/en-us/iaas/Content/generative-ai/pay-on-demand.htm](https://docs.oracle.com/en-us/iaas/Content/generative-ai/pay-on-demand.htm)). The token rates assume 4 characters per token, a planning assumption rather than a listed figure. * AWS Price List API, AmazonBedrockFoundationModels offer: [https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json) * OCI price list API: [https://apexapps.oracle.com/pls/apex/cetools/api/v1/products/](https://apexapps.oracle.com/pls/apex/cetools/api/v1/products/) Cost scope: Modeled target native service and explicitly modeled state allocation; hosting and Tensor9 fees only when supplied. Excluded costs and workload features: * Prompt caching, batch discounts, context tiers, grounding, tools, agents, embeddings, search, guardrails and dedicated inference hardware. * Taxes, discounts, credits and free-tier allowances unless explicitly reflected in edited rates. * Unmodeled dependent services, task execution and operations outside the selected workload family. List prices read: 2026-10-22. Model: `plan-867e301a8e1fa800d4506aa97b8132fa27fdb26a` / `2026-10-22.planning.2`. Curve: `workload-2`. Sources: [AWS Price List API, AmazonBedrockFoundationModels offer](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json), [OCI price list API](https://apexapps.oracle.com/pls/apex/cetools/api/v1/products/). ## Bedrock model inference → OpenAI Chat Completions Origin API: `aws::1.0.0::bedrock`. Target backend: `openai::1.0.0::chat`. ### Bedrock model inference · Input tokens per second Estimated monthly service charges by Input tokens per second. Adapter hosting and Tensor9 fees are excluded. Source: us-east-1; target: global; currency: USD. ![Bedrock model inference · Input tokens per second: monthly service charges by Input tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/275ac832053b693a59a56a72f43aa6bef8db56d2c84bfbbbb4cafc36a4dbb3a7-light.svg) ![Bedrock model inference · Input tokens per second: monthly service charges by Input tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/07c46a5ab69cbeb643008a6132603b4e4100aeb51b949e6bfa1ec9216dc5b73a-dark.svg) Comparison pills describe all sampled scale points; card amounts show the named workload. Costs The Same means within 20% at every point. Costs Depend includes mixed results and unpriced samples. Exact sampled model outputs in USD/month. Difference means target minus origin; a positive difference means the target costs more. | Input tokens per second (tokens/s) | Bedrock (LLM inference) | Target scenario | OpenAI Chat Completions | Difference | | -: | -: | - | -: | -: | | 0 | 375 | OpenAI Chat Completions | 250 | -125 | | 10 | 453.84 | OpenAI Chat Completions | 315.7 | -138.14 | | 100 | 1163.4 | OpenAI Chat Completions | 907 | -256.4 | | 1000 | 8259 | OpenAI Chat Completions | 6820 | -1439 | | 10000 | 79215 | OpenAI Chat Completions | 65950 | -13265 | | 100000 | 788775 | OpenAI Chat Completions | 657250 | -131525 | **Curve assumptions** | Target scenario | Assumption | Value | Unit | | - | - | -: | - | | OpenAI Chat Completions | Origin list price · Input tokens | 0.000003 | USD/Tokens | | OpenAI Chat Completions | Origin list price · Output tokens | 0.000015 | USD/Tokens | | OpenAI Chat Completions | Target / origin input tokens | 1 | multiplier | | OpenAI Chat Completions | Target / origin output tokens | 1 | multiplier | | OpenAI Chat Completions | Target price guess · read | 0 | USD/read | | OpenAI Chat Completions | Target price guess · write | 0 | USD/write | | OpenAI Chat Completions | Target price guess · provider operation | 0 | USD/provider operation | | OpenAI Chat Completions | Target price guess · GiB-month | 0 | USD/GiB-month | | OpenAI Chat Completions | Target price guess · published GiB | 0 | USD/published GiB | | OpenAI Chat Completions | Target price guess · delivered GiB | 0 | USD/delivered GiB | | OpenAI Chat Completions | Target price guess · network GiB | 0 | USD/network GiB | | OpenAI Chat Completions | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | OpenAI Chat Completions | Target price guess · vCPU-hour | 0 | USD/vCPU-hour | | OpenAI Chat Completions | Target price guess · GiB-hour | 0 | USD/GiB-hour | | OpenAI Chat Completions | Target price guess · resource-hour | 0 | USD/resource-hour | | OpenAI Chat Completions | Target price guess · resource-month | 0 | USD/resource-month | | OpenAI Chat Completions | Target list price · input token | 0.0000025 | USD/input token | | OpenAI Chat Completions | Target list price · output token | 0.00001 | USD/output token | | OpenAI Chat Completions | Target monthly base charge | 0 | USD/month | | OpenAI Chat Completions | Approximate target service minimum | 0 | USD/month | | OpenAI Chat Completions | Minimum primary CPU | 2 | vCPU | | OpenAI Chat Completions | Minimum primary RAM | 8 | GiB | | OpenAI Chat Completions | Minimum primary disk | 100 | GiB | | OpenAI Chat Completions | Additional full-price standby copy | 0 | 0 or 1 | | OpenAI Chat Completions | Allocated share of provisioned cost | 1 | fraction | | OpenAI Chat Completions | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | OpenAI Chat Completions | Planning peak / mean work | 2 | multiplier | | OpenAI Chat Completions | Planning primary CPU utilization | 0.7 | fraction | | OpenAI Chat Completions | Resident state reserved in RAM | 0.1 | fraction | | OpenAI Chat Completions | Planning disk utilization | 0.8 | fraction | | OpenAI Chat Completions | Maintenance and log disk reserve | 20 | GiB | | OpenAI Chat Completions | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | OpenAI Chat Completions | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | OpenAI Chat Completions | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | OpenAI Chat Completions | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | OpenAI Chat Completions | Dependent object sink write price guess | 0.000005 | USD/write | * Only this native meter varies; all other source usage and translation assumptions retain the displayed defaults. Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. Other native meters retain their explicitly displayed default quantities. * Planning estimate · editable authored rates and sizing assumptions dated 2026-10-22; not verified provider prices. us-east-1 → global, USD/month. * Source prices are reference charges, not an invoice. Target lines cover selected service usage/configuration only; adapter hosting and Tensor9 fees are excluded. * Explicit tariff samples are not performance measurements or capacity qualification. No interpolation or break-even guarantee is implied. * Variant assumptions show the first axis sample; any axis-dependent configuration is described by this curve's workload note. * Rates are cited provider list prices, editable for discounts, commitments or another region. Workload translation and capacity coefficients remain editable planning assumptions, not benchmark measurements or a commercial quote. * Default-zero rates are explicit assumptions, not evidence that the deployed service or resource is free. * Provisioned-unit minimums use an editable approximate allocation floor. Modeled target service cost is max(usage/allocation, minimum), with billable network, durable staging and dependent sinks added separately; actual RU/TU sizing, scaling increments and throughput remain unqualified. * Durable staging is priced once: provisioned targets include adapter state in their selected allocation; native targets add a separate staging allocation when required. Adapter hosting is a separate allocation and must exclude resources already priced as staging, workers or target capacity. * A declared adapter mapping and cost model do not establish implementation availability or service equivalence. * This is a declared directed mapping; a planning estimate does not establish deployment or request-serving support. * Origin unit prices are cited provider list prices; capacity coefficients and translation assumptions are editable authored planning guesses, not measured capacity or a provider quote. * Origin tariffs are the cited Claude Sonnet 4.6 global-endpoint list prices; target tariffs name no asserted current model SKU. Select model-specific rates before using the estimate. * Input and output meters and token expansion remain independent. A cheaper model is not automatically equivalent in capability or quality. * Input and output token tariffs remain distinct and establish no equivalent output quality or tokenizer. gpt-4o is a nominal choice: the Bedrock adapter's OpenAI and Azure OpenAI examples name it, but the same wire also serves self-hosted vLLM (node cost) and Vertex MaaS models. gpt-4.1 lists 2/8, gpt-5 1.25/10 and gpt-5-mini 0.25/2 per 1M. * AWS Price List API, AmazonBedrockFoundationModels offer: [https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json) * OpenAI API pricing: [https://developers.openai.com/api/docs/pricing](https://developers.openai.com/api/docs/pricing) Cost scope: Modeled target native service and explicitly modeled state allocation; hosting and Tensor9 fees only when supplied. Excluded costs and workload features: * Prompt caching, batch discounts, context tiers, grounding, tools, agents, embeddings, search, guardrails and dedicated inference hardware. * Taxes, discounts, credits and free-tier allowances unless explicitly reflected in edited rates. * Unmodeled dependent services, task execution and operations outside the selected workload family. List prices read: 2026-10-22. Model: `plan-24a146500f73787fcd1f2436716bcced0f4263b3` / `2026-10-22.planning.2`. Curve: `workload-1`. Sources: [AWS Price List API, AmazonBedrockFoundationModels offer](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json), [OpenAI API pricing](https://developers.openai.com/api/docs/pricing). ### Bedrock model inference · Output tokens per second Estimated monthly service charges by Output tokens per second. Adapter hosting and Tensor9 fees are excluded. Source: us-east-1; target: global; currency: USD. ![Bedrock model inference · Output tokens per second: monthly service charges by Output tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/b733c83af117bab487615c7f9fc9107f9b22a4940b8fe7eabe102762acc07d7d-light.svg) ![Bedrock model inference · Output tokens per second: monthly service charges by Output tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/1b74ad590ac343953d2c4917ef6c9aa1f6c9992ea5068630d76c4b7b5defbcfe-dark.svg) Comparison pills describe all sampled scale points; card amounts show the named workload. Costs The Same means within 20% at every point. Costs Depend includes mixed results and unpriced samples. Exact sampled model outputs in USD/month. Difference means target minus origin; a positive difference means the target costs more. | Output tokens per second (tokens/s) | Bedrock (LLM inference) | Target scenario | OpenAI Chat Completions | Difference | | -: | -: | - | -: | -: | | 0 | 300 | OpenAI Chat Completions | 250 | -50 | | 10 | 694.2 | OpenAI Chat Completions | 512.8 | -181.4 | | 100 | 4242 | OpenAI Chat Completions | 2878 | -1364 | | 1000 | 39720 | OpenAI Chat Completions | 26530 | -13190 | | 10000 | 394500 | OpenAI Chat Completions | 263050 | -131450 | | 100000 | 3942300 | OpenAI Chat Completions | 2628250 | -1314050 | **Curve assumptions** | Target scenario | Assumption | Value | Unit | | - | - | -: | - | | OpenAI Chat Completions | Origin list price · Input tokens | 0.000003 | USD/Tokens | | OpenAI Chat Completions | Origin list price · Output tokens | 0.000015 | USD/Tokens | | OpenAI Chat Completions | Target / origin input tokens | 1 | multiplier | | OpenAI Chat Completions | Target / origin output tokens | 1 | multiplier | | OpenAI Chat Completions | Target price guess · read | 0 | USD/read | | OpenAI Chat Completions | Target price guess · write | 0 | USD/write | | OpenAI Chat Completions | Target price guess · provider operation | 0 | USD/provider operation | | OpenAI Chat Completions | Target price guess · GiB-month | 0 | USD/GiB-month | | OpenAI Chat Completions | Target price guess · published GiB | 0 | USD/published GiB | | OpenAI Chat Completions | Target price guess · delivered GiB | 0 | USD/delivered GiB | | OpenAI Chat Completions | Target price guess · network GiB | 0 | USD/network GiB | | OpenAI Chat Completions | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | OpenAI Chat Completions | Target price guess · vCPU-hour | 0 | USD/vCPU-hour | | OpenAI Chat Completions | Target price guess · GiB-hour | 0 | USD/GiB-hour | | OpenAI Chat Completions | Target price guess · resource-hour | 0 | USD/resource-hour | | OpenAI Chat Completions | Target price guess · resource-month | 0 | USD/resource-month | | OpenAI Chat Completions | Target list price · input token | 0.0000025 | USD/input token | | OpenAI Chat Completions | Target list price · output token | 0.00001 | USD/output token | | OpenAI Chat Completions | Target monthly base charge | 0 | USD/month | | OpenAI Chat Completions | Approximate target service minimum | 0 | USD/month | | OpenAI Chat Completions | Minimum primary CPU | 2 | vCPU | | OpenAI Chat Completions | Minimum primary RAM | 8 | GiB | | OpenAI Chat Completions | Minimum primary disk | 100 | GiB | | OpenAI Chat Completions | Additional full-price standby copy | 0 | 0 or 1 | | OpenAI Chat Completions | Allocated share of provisioned cost | 1 | fraction | | OpenAI Chat Completions | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | OpenAI Chat Completions | Planning peak / mean work | 2 | multiplier | | OpenAI Chat Completions | Planning primary CPU utilization | 0.7 | fraction | | OpenAI Chat Completions | Resident state reserved in RAM | 0.1 | fraction | | OpenAI Chat Completions | Planning disk utilization | 0.8 | fraction | | OpenAI Chat Completions | Maintenance and log disk reserve | 20 | GiB | | OpenAI Chat Completions | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | OpenAI Chat Completions | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | OpenAI Chat Completions | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | OpenAI Chat Completions | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | OpenAI Chat Completions | Dependent object sink write price guess | 0.000005 | USD/write | * Only this native meter varies; all other source usage and translation assumptions retain the displayed defaults. Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. Other native meters retain their explicitly displayed default quantities. * Planning estimate · editable authored rates and sizing assumptions dated 2026-10-22; not verified provider prices. us-east-1 → global, USD/month. * Source prices are reference charges, not an invoice. Target lines cover selected service usage/configuration only; adapter hosting and Tensor9 fees are excluded. * Explicit tariff samples are not performance measurements or capacity qualification. No interpolation or break-even guarantee is implied. * Variant assumptions show the first axis sample; any axis-dependent configuration is described by this curve's workload note. * Rates are cited provider list prices, editable for discounts, commitments or another region. Workload translation and capacity coefficients remain editable planning assumptions, not benchmark measurements or a commercial quote. * Default-zero rates are explicit assumptions, not evidence that the deployed service or resource is free. * Provisioned-unit minimums use an editable approximate allocation floor. Modeled target service cost is max(usage/allocation, minimum), with billable network, durable staging and dependent sinks added separately; actual RU/TU sizing, scaling increments and throughput remain unqualified. * Durable staging is priced once: provisioned targets include adapter state in their selected allocation; native targets add a separate staging allocation when required. Adapter hosting is a separate allocation and must exclude resources already priced as staging, workers or target capacity. * A declared adapter mapping and cost model do not establish implementation availability or service equivalence. * This is a declared directed mapping; a planning estimate does not establish deployment or request-serving support. * Origin unit prices are cited provider list prices; capacity coefficients and translation assumptions are editable authored planning guesses, not measured capacity or a provider quote. * Origin tariffs are the cited Claude Sonnet 4.6 global-endpoint list prices; target tariffs name no asserted current model SKU. Select model-specific rates before using the estimate. * Input and output meters and token expansion remain independent. A cheaper model is not automatically equivalent in capability or quality. * Input and output token tariffs remain distinct and establish no equivalent output quality or tokenizer. gpt-4o is a nominal choice: the Bedrock adapter's OpenAI and Azure OpenAI examples name it, but the same wire also serves self-hosted vLLM (node cost) and Vertex MaaS models. gpt-4.1 lists 2/8, gpt-5 1.25/10 and gpt-5-mini 0.25/2 per 1M. * AWS Price List API, AmazonBedrockFoundationModels offer: [https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json) * OpenAI API pricing: [https://developers.openai.com/api/docs/pricing](https://developers.openai.com/api/docs/pricing) Cost scope: Modeled target native service and explicitly modeled state allocation; hosting and Tensor9 fees only when supplied. Excluded costs and workload features: * Prompt caching, batch discounts, context tiers, grounding, tools, agents, embeddings, search, guardrails and dedicated inference hardware. * Taxes, discounts, credits and free-tier allowances unless explicitly reflected in edited rates. * Unmodeled dependent services, task execution and operations outside the selected workload family. List prices read: 2026-10-22. Model: `plan-24a146500f73787fcd1f2436716bcced0f4263b3` / `2026-10-22.planning.2`. Curve: `workload-2`. Sources: [AWS Price List API, AmazonBedrockFoundationModels offer](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json), [OpenAI API pricing](https://developers.openai.com/api/docs/pricing). ## Bedrock model inference → Scaleway Managed Inference Origin API: `aws::1.0.0::bedrock`. Target backend: `scaleway::1.0.0::inference`. ### Bedrock model inference · Input tokens per second Estimated monthly service charges by Input tokens per second. Adapter hosting and Tensor9 fees are excluded. Source: us-east-1; target: Explicit target planning region; currency: USD. ![Bedrock model inference · Input tokens per second: monthly service charges by Input tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/d4ab704c3deb2ce81739113a68d79095f2fd180fc262fe81c8dc9baf1b628ebf-light.svg) ![Bedrock model inference · Input tokens per second: monthly service charges by Input tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/282b8504bdd79a9f4ae201931dd275b472623960f160ef4ab9c8019c0055e3f7-dark.svg) Comparison pills describe all sampled scale points; card amounts show the named workload. Costs The Same means within 20% at every point. Costs Depend includes mixed results and unpriced samples. Exact sampled model outputs in USD/month. Difference means target minus origin; a positive difference means the target costs more. Estimated allocation from planning assumptions; not benchmark calibrated. Lines connect evaluated samples; intermediate capacity and prices are not inferred. Unpriced means no target quote or cost difference is available within the sizing policy. | Input tokens per second (tokens/s) | Bedrock (LLM inference) | Target scenario | Scaleway Managed Inference | Difference | | -: | -: | - | -: | -: | | 0 | 375 | Scaleway Managed Inference (one copy) | Unpriced | Unavailable | | 0 | 375 | Scaleway Managed Inference (two copies) | Unpriced | Unavailable | | 10 | 453.84 | Scaleway Managed Inference (one copy) | Unpriced | Unavailable | | 10 | 453.84 | Scaleway Managed Inference (two copies) | Unpriced | Unavailable | | 100 | 1163.4 | Scaleway Managed Inference (one copy) | Unpriced | Unavailable | | 100 | 1163.4 | Scaleway Managed Inference (two copies) | Unpriced | Unavailable | | 1000 | 8259 | Scaleway Managed Inference (one copy) | Unpriced | Unavailable | | 1000 | 8259 | Scaleway Managed Inference (two copies) | Unpriced | Unavailable | | 10000 | 79215 | Scaleway Managed Inference (one copy) | Unpriced | Unavailable | | 10000 | 79215 | Scaleway Managed Inference (two copies) | Unpriced | Unavailable | | 100000 | 788775 | Scaleway Managed Inference (one copy) | Unpriced | Unavailable | | 100000 | 788775 | Scaleway Managed Inference (two copies) | Unpriced | Unavailable | **Estimated allocation from planning assumptions; not benchmark calibrated.** The sizing policy grows compute with translated work and peak headroom, memory with compute and working set, and disk with resident working state and reserve. Monthly usage is distinct from resident data; each workload's assumptions define the translation. HA changes pricing, not this policy's throughput estimate. Unpriced workloads exceed the policy's configuration range; no silent vertical cap or automatic sharding is assumed. | Sample / scenario | Basis | Mean / peak work/s | Required vCPU | Allocated vCPU | Memory GiB | SSD GiB | Resident GiB | Policy | Constraints | | - | - | -: | -: | -: | -: | -: | -: | - | - | | 0 tokens/s / Scaleway Managed Inference (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 0 tokens/s / Scaleway Managed Inference (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10 tokens/s / Scaleway Managed Inference (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10 tokens/s / Scaleway Managed Inference (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100 tokens/s / Scaleway Managed Inference (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100 tokens/s / Scaleway Managed Inference (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 1000 tokens/s / Scaleway Managed Inference (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 1000 tokens/s / Scaleway Managed Inference (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10000 tokens/s / Scaleway Managed Inference (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10000 tokens/s / Scaleway Managed Inference (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100000 tokens/s / Scaleway Managed Inference (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100000 tokens/s / Scaleway Managed Inference (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | * Row-store throughput (capacityOpsPerVcpuSecond, default 1500) is measured-by-tensor9 as a lower bound: the DynamoDB adapter on Cloud SQL db-custom-2-7680 (2 vCPU, PostgreSQL 16, us-west1) served 2,974 ops/s of YCSB-A at 16 threads without recorded saturation (src/main/kotlin/t9/prime/equiv/generated/eval/CloudsqlPostgresqlYcsbAT16GoYcsbV1\_0\_0.kt). Published CPU-bound pgbench reaches 3,101 read-write transactions and 20,289 point reads per vCPU-second (AWS z1d.2xlarge, PostgreSQL 14.5), so actual capacity is likely higher. Reads and writes share one coefficient: at the same adapter version, read-only YCSB-C was not faster than 50%-update YCSB-A. * Log-broker throughput (capacityStreamMiBPerVcpuSecond 20 at 0.5 utilization, 4 GiB RAM per vCPU) is provider-documented by Google Managed Service for Apache Kafka sizing ([https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size](https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size)): write-equivalent bytes are replicated writes plus a quarter of reads. Producer batches below 10 KB lower per-CPU throughput; that penalty is not modeled. * Disk utilization 0.8 is provider-documented headroom below Cloud SQL automatic storage increase, which triggers near 91-95% full ([https://docs.cloud.google.com/sql/docs/postgres/instance-settings](https://docs.cloud.google.com/sql/docs/postgres/instance-settings)). The 20 GiB reserve exceeds PostgreSQL's 1 GB default max\_wal\_size; the remainder of that reserve is unmeasured. * CPU tiers, 4 GiB per vCPU, at most 6.5 GiB per vCPU, 96 vCPU and 65,536 GiB disk follow Cloud SQL Enterprise custom machine limits (provider-documented, same page); 624 GiB is 96 × 6.5. Targets billed per node, unit or throughput unit are sized by their backend instead. * No public evidence: peak/mean work (2) and the RAM working-set fraction (0.1). Proposed measurements: the p99 one-minute rate over the monthly mean from customer CloudWatch metrics; the buffer hit ratio while stepping RAM under a fixed YCSB-B load. * Stored GiB already includes the origin family's physical footprint assumptions; no second storage amplification is applied. * The operations/vCPU coefficient supplies the compute requirement alongside explicit resource-hour minima. * Copies and allocation fractions change cost only; they do not increase modeled primary capacity. **Curve assumptions** | Target scenario | Assumption | Value | Unit | | - | - | -: | - | | Scaleway Managed Inference (one copy) | Origin list price · Input tokens | 0.000003 | USD/Tokens | | Scaleway Managed Inference (one copy) | Origin list price · Output tokens | 0.000015 | USD/Tokens | | Scaleway Managed Inference (one copy) | Target / origin input tokens | 1 | multiplier | | Scaleway Managed Inference (one copy) | Target / origin output tokens | 1 | multiplier | | Scaleway Managed Inference (one copy) | Target price guess · read | 0 | USD/read | | Scaleway Managed Inference (one copy) | Target price guess · write | 0 | USD/write | | Scaleway Managed Inference (one copy) | Target price guess · provider operation | 0 | USD/provider operation | | Scaleway Managed Inference (one copy) | Target price guess · GiB-month | 0.17 | USD/GiB-month | | Scaleway Managed Inference (one copy) | Target price guess · published GiB | 0 | USD/published GiB | | Scaleway Managed Inference (one copy) | Target price guess · delivered GiB | 0 | USD/delivered GiB | | Scaleway Managed Inference (one copy) | Target price guess · network GiB | 0.09 | USD/network GiB | | Scaleway Managed Inference (one copy) | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | Scaleway Managed Inference (one copy) | Target price guess · vCPU-hour | 0.05 | USD/vCPU-hour | | Scaleway Managed Inference (one copy) | Target price guess · GiB-hour | 0.006 | USD/GiB-hour | | Scaleway Managed Inference (one copy) | Target price guess · resource-hour | 0 | USD/resource-hour | | Scaleway Managed Inference (one copy) | Target price guess · resource-month | 0 | USD/resource-month | | Scaleway Managed Inference (one copy) | Target price guess · input token | 0 | USD/input token | | Scaleway Managed Inference (one copy) | Target price guess · output token | 0 | USD/output token | | Scaleway Managed Inference (one copy) | Target monthly base charge | 0 | USD/month | | Scaleway Managed Inference (one copy) | Approximate target service minimum | 0 | USD/month | | Scaleway Managed Inference (one copy) | Minimum primary CPU | 2 | vCPU | | Scaleway Managed Inference (one copy) | Minimum primary RAM | 8 | GiB | | Scaleway Managed Inference (one copy) | Minimum primary disk | 100 | GiB | | Scaleway Managed Inference (one copy) | Additional full-price standby copy | 0 | 0 or 1 | | Scaleway Managed Inference (one copy) | Allocated share of provisioned cost | 1 | fraction | | Scaleway Managed Inference (one copy) | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | Scaleway Managed Inference (one copy) | Planning peak / mean work | 2 | multiplier | | Scaleway Managed Inference (one copy) | Planning primary CPU utilization | 0.7 | fraction | | Scaleway Managed Inference (one copy) | Resident state reserved in RAM | 0.1 | fraction | | Scaleway Managed Inference (one copy) | Planning disk utilization | 0.8 | fraction | | Scaleway Managed Inference (one copy) | Maintenance and log disk reserve | 20 | GiB | | Scaleway Managed Inference (one copy) | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | Scaleway Managed Inference (one copy) | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | Scaleway Managed Inference (one copy) | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | Scaleway Managed Inference (one copy) | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | Scaleway Managed Inference (one copy) | Dependent object sink write price guess | 0.000005 | USD/write | | Scaleway Managed Inference (two copies) | Origin list price · Input tokens | 0.000003 | USD/Tokens | | Scaleway Managed Inference (two copies) | Origin list price · Output tokens | 0.000015 | USD/Tokens | | Scaleway Managed Inference (two copies) | Target / origin input tokens | 1 | multiplier | | Scaleway Managed Inference (two copies) | Target / origin output tokens | 1 | multiplier | | Scaleway Managed Inference (two copies) | Target price guess · read | 0 | USD/read | | Scaleway Managed Inference (two copies) | Target price guess · write | 0 | USD/write | | Scaleway Managed Inference (two copies) | Target price guess · provider operation | 0 | USD/provider operation | | Scaleway Managed Inference (two copies) | Target price guess · GiB-month | 0.17 | USD/GiB-month | | Scaleway Managed Inference (two copies) | Target price guess · published GiB | 0 | USD/published GiB | | Scaleway Managed Inference (two copies) | Target price guess · delivered GiB | 0 | USD/delivered GiB | | Scaleway Managed Inference (two copies) | Target price guess · network GiB | 0.09 | USD/network GiB | | Scaleway Managed Inference (two copies) | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | Scaleway Managed Inference (two copies) | Target price guess · vCPU-hour | 0.05 | USD/vCPU-hour | | Scaleway Managed Inference (two copies) | Target price guess · GiB-hour | 0.006 | USD/GiB-hour | | Scaleway Managed Inference (two copies) | Target price guess · resource-hour | 0 | USD/resource-hour | | Scaleway Managed Inference (two copies) | Target price guess · resource-month | 0 | USD/resource-month | | Scaleway Managed Inference (two copies) | Target price guess · input token | 0 | USD/input token | | Scaleway Managed Inference (two copies) | Target price guess · output token | 0 | USD/output token | | Scaleway Managed Inference (two copies) | Target monthly base charge | 0 | USD/month | | Scaleway Managed Inference (two copies) | Approximate target service minimum | 0 | USD/month | | Scaleway Managed Inference (two copies) | Minimum primary CPU | 2 | vCPU | | Scaleway Managed Inference (two copies) | Minimum primary RAM | 8 | GiB | | Scaleway Managed Inference (two copies) | Minimum primary disk | 100 | GiB | | Scaleway Managed Inference (two copies) | Additional full-price standby copy | 1 | 0 or 1 | | Scaleway Managed Inference (two copies) | Allocated share of provisioned cost | 1 | fraction | | Scaleway Managed Inference (two copies) | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | Scaleway Managed Inference (two copies) | Planning peak / mean work | 2 | multiplier | | Scaleway Managed Inference (two copies) | Planning primary CPU utilization | 0.7 | fraction | | Scaleway Managed Inference (two copies) | Resident state reserved in RAM | 0.1 | fraction | | Scaleway Managed Inference (two copies) | Planning disk utilization | 0.8 | fraction | | Scaleway Managed Inference (two copies) | Maintenance and log disk reserve | 20 | GiB | | Scaleway Managed Inference (two copies) | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | Scaleway Managed Inference (two copies) | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | Scaleway Managed Inference (two copies) | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | Scaleway Managed Inference (two copies) | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | Scaleway Managed Inference (two copies) | Dependent object sink write price guess | 0.000005 | USD/write | * Only this native meter varies; all other source usage and translation assumptions retain the displayed defaults. Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. Other native meters retain their explicitly displayed default quantities. * Planning estimate · editable authored rates and sizing assumptions dated 2026-10-22; not verified provider prices. us-east-1 → Explicit target planning region, USD/month. * Source prices are reference charges, not an invoice. Target lines cover selected service usage/configuration only; adapter hosting and Tensor9 fees are excluded. * Explicit tariff samples are not performance measurements or capacity qualification. No interpolation or break-even guarantee is implied. * Variant assumptions show the first axis sample; any axis-dependent configuration is described by this curve's workload note. * Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. * Default-zero rates are explicit assumptions, not evidence that the deployed service or resource is free. * Provisioned-unit minimums use an editable approximate allocation floor. Modeled target service cost is max(usage/allocation, minimum), with billable network, durable staging and dependent sinks added separately; actual RU/TU sizing, scaling increments and throughput remain unqualified. * Durable staging is priced once: provisioned targets include adapter state in their selected allocation; native targets add a separate staging allocation when required. Adapter hosting is a separate allocation and must exclude resources already priced as staging, workers or target capacity. * A declared adapter mapping and cost model do not establish implementation availability or service equivalence. * This is a declared directed mapping; a planning estimate does not establish deployment or request-serving support. * Origin unit prices are cited provider list prices; capacity coefficients and translation assumptions are editable authored planning guesses, not measured capacity or a provider quote. * Origin tariffs are the cited Claude Sonnet 4.6 global-endpoint list prices; target tariffs name no asserted current model SKU. Select model-specific rates before using the estimate. * Input and output meters and token expansion remain independent. A cheaper model is not automatically equivalent in capability or quality. * All target rates are editable authored planning guesses, not verified current list prices, customer discounts or benchmark evidence. * Dedicated accelerator/model-fit target: the Bedrock family returns this direction as Unpriced because the serving GPU or model deployment is not selected, so these generic rates are not shown as list prices. Scaleway Managed Inference lists EUR 0.93/hour for an L4-1-24G node ([https://www.scaleway.com/en/pricing/model-as-a-service/](https://www.scaleway.com/en/pricing/model-as-a-service/)); the Azure RAI policy has no meter of its own, its tokens bill on the Azure OpenAI deployment and knowledge bases on Azure AI Search (Basic USD 0.101/hour); the Tensor9 in-cluster model service needs GPU node pricing. Read 2026-10-22. * Row-store throughput (capacityOpsPerVcpuSecond, default 1500) is measured-by-tensor9 as a lower bound: the DynamoDB adapter on Cloud SQL db-custom-2-7680 (2 vCPU, PostgreSQL 16, us-west1) served 2,974 ops/s of YCSB-A at 16 threads without recorded saturation (src/main/kotlin/t9/prime/equiv/generated/eval/CloudsqlPostgresqlYcsbAT16GoYcsbV1\_0\_0.kt). Published CPU-bound pgbench reaches 3,101 read-write transactions and 20,289 point reads per vCPU-second (AWS z1d.2xlarge, PostgreSQL 14.5), so actual capacity is likely higher. Reads and writes share one coefficient: at the same adapter version, read-only YCSB-C was not faster than 50%-update YCSB-A. * Log-broker throughput (capacityStreamMiBPerVcpuSecond 20 at 0.5 utilization, 4 GiB RAM per vCPU) is provider-documented by Google Managed Service for Apache Kafka sizing ([https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size](https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size)): write-equivalent bytes are replicated writes plus a quarter of reads. Producer batches below 10 KB lower per-CPU throughput; that penalty is not modeled. * Disk utilization 0.8 is provider-documented headroom below Cloud SQL automatic storage increase, which triggers near 91-95% full ([https://docs.cloud.google.com/sql/docs/postgres/instance-settings](https://docs.cloud.google.com/sql/docs/postgres/instance-settings)). The 20 GiB reserve exceeds PostgreSQL's 1 GB default max\_wal\_size; the remainder of that reserve is unmeasured. * CPU tiers, 4 GiB per vCPU, at most 6.5 GiB per vCPU, 96 vCPU and 65,536 GiB disk follow Cloud SQL Enterprise custom machine limits (provider-documented, same page); 624 GiB is 96 × 6.5. Targets billed per node, unit or throughput unit are sized by their backend instead. * No public evidence: peak/mean work (2) and the RAM working-set fraction (0.1). Proposed measurements: the p99 one-minute rate over the monthly mean from customer CloudWatch metrics; the buffer hit ratio while stepping RAM under a fixed YCSB-B load. * Stored GiB already includes the origin family's physical footprint assumptions; no second storage amplification is applied. * The operations/vCPU coefficient supplies the compute requirement alongside explicit resource-hour minima. * Copies and allocation fractions change cost only; they do not increase modeled primary capacity. * AWS Price List API, AmazonBedrockFoundationModels offer: [https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json) Cost scope: Modeled target native service and explicitly modeled state allocation; hosting and Tensor9 fees only when supplied. Excluded costs and workload features: * Prompt caching, batch discounts, context tiers, grounding, tools, agents, embeddings, search, guardrails and dedicated inference hardware. * Taxes, discounts, credits and free-tier allowances unless explicitly reflected in edited rates. * Unmodeled dependent services, task execution and operations outside the selected workload family. List prices read: 2026-10-22. Model: `plan-a6f9604e9c2919f36a8bcff1239ddada7e8d8663` / `2026-10-22.planning.2`. Curve: `workload-1`. Sources: [AWS Price List API, AmazonBedrockFoundationModels offer](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json). ### Bedrock model inference · Output tokens per second Estimated monthly service charges by Output tokens per second. Adapter hosting and Tensor9 fees are excluded. Source: us-east-1; target: Explicit target planning region; currency: USD. ![Bedrock model inference · Output tokens per second: monthly service charges by Output tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/ed5394c234fe3720de8dfd619d8011b99f485bec95afc418d291421c8afc69b1-light.svg) ![Bedrock model inference · Output tokens per second: monthly service charges by Output tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/17931b031f64d0042414e71763d8e6df63e76fc2a08fd8fcf4547f083ecd6dc6-dark.svg) Comparison pills describe all sampled scale points; card amounts show the named workload. Costs The Same means within 20% at every point. Costs Depend includes mixed results and unpriced samples. Exact sampled model outputs in USD/month. Difference means target minus origin; a positive difference means the target costs more. Estimated allocation from planning assumptions; not benchmark calibrated. Lines connect evaluated samples; intermediate capacity and prices are not inferred. Unpriced means no target quote or cost difference is available within the sizing policy. | Output tokens per second (tokens/s) | Bedrock (LLM inference) | Target scenario | Scaleway Managed Inference | Difference | | -: | -: | - | -: | -: | | 0 | 300 | Scaleway Managed Inference (one copy) | Unpriced | Unavailable | | 0 | 300 | Scaleway Managed Inference (two copies) | Unpriced | Unavailable | | 10 | 694.2 | Scaleway Managed Inference (one copy) | Unpriced | Unavailable | | 10 | 694.2 | Scaleway Managed Inference (two copies) | Unpriced | Unavailable | | 100 | 4242 | Scaleway Managed Inference (one copy) | Unpriced | Unavailable | | 100 | 4242 | Scaleway Managed Inference (two copies) | Unpriced | Unavailable | | 1000 | 39720 | Scaleway Managed Inference (one copy) | Unpriced | Unavailable | | 1000 | 39720 | Scaleway Managed Inference (two copies) | Unpriced | Unavailable | | 10000 | 394500 | Scaleway Managed Inference (one copy) | Unpriced | Unavailable | | 10000 | 394500 | Scaleway Managed Inference (two copies) | Unpriced | Unavailable | | 100000 | 3942300 | Scaleway Managed Inference (one copy) | Unpriced | Unavailable | | 100000 | 3942300 | Scaleway Managed Inference (two copies) | Unpriced | Unavailable | **Estimated allocation from planning assumptions; not benchmark calibrated.** The sizing policy grows compute with translated work and peak headroom, memory with compute and working set, and disk with resident working state and reserve. Monthly usage is distinct from resident data; each workload's assumptions define the translation. HA changes pricing, not this policy's throughput estimate. Unpriced workloads exceed the policy's configuration range; no silent vertical cap or automatic sharding is assumed. | Sample / scenario | Basis | Mean / peak work/s | Required vCPU | Allocated vCPU | Memory GiB | SSD GiB | Resident GiB | Policy | Constraints | | - | - | -: | -: | -: | -: | -: | -: | - | - | | 0 tokens/s / Scaleway Managed Inference (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 0 tokens/s / Scaleway Managed Inference (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10 tokens/s / Scaleway Managed Inference (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10 tokens/s / Scaleway Managed Inference (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100 tokens/s / Scaleway Managed Inference (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100 tokens/s / Scaleway Managed Inference (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 1000 tokens/s / Scaleway Managed Inference (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 1000 tokens/s / Scaleway Managed Inference (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10000 tokens/s / Scaleway Managed Inference (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10000 tokens/s / Scaleway Managed Inference (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100000 tokens/s / Scaleway Managed Inference (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100000 tokens/s / Scaleway Managed Inference (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | * Row-store throughput (capacityOpsPerVcpuSecond, default 1500) is measured-by-tensor9 as a lower bound: the DynamoDB adapter on Cloud SQL db-custom-2-7680 (2 vCPU, PostgreSQL 16, us-west1) served 2,974 ops/s of YCSB-A at 16 threads without recorded saturation (src/main/kotlin/t9/prime/equiv/generated/eval/CloudsqlPostgresqlYcsbAT16GoYcsbV1\_0\_0.kt). Published CPU-bound pgbench reaches 3,101 read-write transactions and 20,289 point reads per vCPU-second (AWS z1d.2xlarge, PostgreSQL 14.5), so actual capacity is likely higher. Reads and writes share one coefficient: at the same adapter version, read-only YCSB-C was not faster than 50%-update YCSB-A. * Log-broker throughput (capacityStreamMiBPerVcpuSecond 20 at 0.5 utilization, 4 GiB RAM per vCPU) is provider-documented by Google Managed Service for Apache Kafka sizing ([https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size](https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size)): write-equivalent bytes are replicated writes plus a quarter of reads. Producer batches below 10 KB lower per-CPU throughput; that penalty is not modeled. * Disk utilization 0.8 is provider-documented headroom below Cloud SQL automatic storage increase, which triggers near 91-95% full ([https://docs.cloud.google.com/sql/docs/postgres/instance-settings](https://docs.cloud.google.com/sql/docs/postgres/instance-settings)). The 20 GiB reserve exceeds PostgreSQL's 1 GB default max\_wal\_size; the remainder of that reserve is unmeasured. * CPU tiers, 4 GiB per vCPU, at most 6.5 GiB per vCPU, 96 vCPU and 65,536 GiB disk follow Cloud SQL Enterprise custom machine limits (provider-documented, same page); 624 GiB is 96 × 6.5. Targets billed per node, unit or throughput unit are sized by their backend instead. * No public evidence: peak/mean work (2) and the RAM working-set fraction (0.1). Proposed measurements: the p99 one-minute rate over the monthly mean from customer CloudWatch metrics; the buffer hit ratio while stepping RAM under a fixed YCSB-B load. * Stored GiB already includes the origin family's physical footprint assumptions; no second storage amplification is applied. * The operations/vCPU coefficient supplies the compute requirement alongside explicit resource-hour minima. * Copies and allocation fractions change cost only; they do not increase modeled primary capacity. **Curve assumptions** | Target scenario | Assumption | Value | Unit | | - | - | -: | - | | Scaleway Managed Inference (one copy) | Origin list price · Input tokens | 0.000003 | USD/Tokens | | Scaleway Managed Inference (one copy) | Origin list price · Output tokens | 0.000015 | USD/Tokens | | Scaleway Managed Inference (one copy) | Target / origin input tokens | 1 | multiplier | | Scaleway Managed Inference (one copy) | Target / origin output tokens | 1 | multiplier | | Scaleway Managed Inference (one copy) | Target price guess · read | 0 | USD/read | | Scaleway Managed Inference (one copy) | Target price guess · write | 0 | USD/write | | Scaleway Managed Inference (one copy) | Target price guess · provider operation | 0 | USD/provider operation | | Scaleway Managed Inference (one copy) | Target price guess · GiB-month | 0.17 | USD/GiB-month | | Scaleway Managed Inference (one copy) | Target price guess · published GiB | 0 | USD/published GiB | | Scaleway Managed Inference (one copy) | Target price guess · delivered GiB | 0 | USD/delivered GiB | | Scaleway Managed Inference (one copy) | Target price guess · network GiB | 0.09 | USD/network GiB | | Scaleway Managed Inference (one copy) | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | Scaleway Managed Inference (one copy) | Target price guess · vCPU-hour | 0.05 | USD/vCPU-hour | | Scaleway Managed Inference (one copy) | Target price guess · GiB-hour | 0.006 | USD/GiB-hour | | Scaleway Managed Inference (one copy) | Target price guess · resource-hour | 0 | USD/resource-hour | | Scaleway Managed Inference (one copy) | Target price guess · resource-month | 0 | USD/resource-month | | Scaleway Managed Inference (one copy) | Target price guess · input token | 0 | USD/input token | | Scaleway Managed Inference (one copy) | Target price guess · output token | 0 | USD/output token | | Scaleway Managed Inference (one copy) | Target monthly base charge | 0 | USD/month | | Scaleway Managed Inference (one copy) | Approximate target service minimum | 0 | USD/month | | Scaleway Managed Inference (one copy) | Minimum primary CPU | 2 | vCPU | | Scaleway Managed Inference (one copy) | Minimum primary RAM | 8 | GiB | | Scaleway Managed Inference (one copy) | Minimum primary disk | 100 | GiB | | Scaleway Managed Inference (one copy) | Additional full-price standby copy | 0 | 0 or 1 | | Scaleway Managed Inference (one copy) | Allocated share of provisioned cost | 1 | fraction | | Scaleway Managed Inference (one copy) | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | Scaleway Managed Inference (one copy) | Planning peak / mean work | 2 | multiplier | | Scaleway Managed Inference (one copy) | Planning primary CPU utilization | 0.7 | fraction | | Scaleway Managed Inference (one copy) | Resident state reserved in RAM | 0.1 | fraction | | Scaleway Managed Inference (one copy) | Planning disk utilization | 0.8 | fraction | | Scaleway Managed Inference (one copy) | Maintenance and log disk reserve | 20 | GiB | | Scaleway Managed Inference (one copy) | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | Scaleway Managed Inference (one copy) | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | Scaleway Managed Inference (one copy) | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | Scaleway Managed Inference (one copy) | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | Scaleway Managed Inference (one copy) | Dependent object sink write price guess | 0.000005 | USD/write | | Scaleway Managed Inference (two copies) | Origin list price · Input tokens | 0.000003 | USD/Tokens | | Scaleway Managed Inference (two copies) | Origin list price · Output tokens | 0.000015 | USD/Tokens | | Scaleway Managed Inference (two copies) | Target / origin input tokens | 1 | multiplier | | Scaleway Managed Inference (two copies) | Target / origin output tokens | 1 | multiplier | | Scaleway Managed Inference (two copies) | Target price guess · read | 0 | USD/read | | Scaleway Managed Inference (two copies) | Target price guess · write | 0 | USD/write | | Scaleway Managed Inference (two copies) | Target price guess · provider operation | 0 | USD/provider operation | | Scaleway Managed Inference (two copies) | Target price guess · GiB-month | 0.17 | USD/GiB-month | | Scaleway Managed Inference (two copies) | Target price guess · published GiB | 0 | USD/published GiB | | Scaleway Managed Inference (two copies) | Target price guess · delivered GiB | 0 | USD/delivered GiB | | Scaleway Managed Inference (two copies) | Target price guess · network GiB | 0.09 | USD/network GiB | | Scaleway Managed Inference (two copies) | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | Scaleway Managed Inference (two copies) | Target price guess · vCPU-hour | 0.05 | USD/vCPU-hour | | Scaleway Managed Inference (two copies) | Target price guess · GiB-hour | 0.006 | USD/GiB-hour | | Scaleway Managed Inference (two copies) | Target price guess · resource-hour | 0 | USD/resource-hour | | Scaleway Managed Inference (two copies) | Target price guess · resource-month | 0 | USD/resource-month | | Scaleway Managed Inference (two copies) | Target price guess · input token | 0 | USD/input token | | Scaleway Managed Inference (two copies) | Target price guess · output token | 0 | USD/output token | | Scaleway Managed Inference (two copies) | Target monthly base charge | 0 | USD/month | | Scaleway Managed Inference (two copies) | Approximate target service minimum | 0 | USD/month | | Scaleway Managed Inference (two copies) | Minimum primary CPU | 2 | vCPU | | Scaleway Managed Inference (two copies) | Minimum primary RAM | 8 | GiB | | Scaleway Managed Inference (two copies) | Minimum primary disk | 100 | GiB | | Scaleway Managed Inference (two copies) | Additional full-price standby copy | 1 | 0 or 1 | | Scaleway Managed Inference (two copies) | Allocated share of provisioned cost | 1 | fraction | | Scaleway Managed Inference (two copies) | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | Scaleway Managed Inference (two copies) | Planning peak / mean work | 2 | multiplier | | Scaleway Managed Inference (two copies) | Planning primary CPU utilization | 0.7 | fraction | | Scaleway Managed Inference (two copies) | Resident state reserved in RAM | 0.1 | fraction | | Scaleway Managed Inference (two copies) | Planning disk utilization | 0.8 | fraction | | Scaleway Managed Inference (two copies) | Maintenance and log disk reserve | 20 | GiB | | Scaleway Managed Inference (two copies) | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | Scaleway Managed Inference (two copies) | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | Scaleway Managed Inference (two copies) | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | Scaleway Managed Inference (two copies) | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | Scaleway Managed Inference (two copies) | Dependent object sink write price guess | 0.000005 | USD/write | * Only this native meter varies; all other source usage and translation assumptions retain the displayed defaults. Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. Other native meters retain their explicitly displayed default quantities. * Planning estimate · editable authored rates and sizing assumptions dated 2026-10-22; not verified provider prices. us-east-1 → Explicit target planning region, USD/month. * Source prices are reference charges, not an invoice. Target lines cover selected service usage/configuration only; adapter hosting and Tensor9 fees are excluded. * Explicit tariff samples are not performance measurements or capacity qualification. No interpolation or break-even guarantee is implied. * Variant assumptions show the first axis sample; any axis-dependent configuration is described by this curve's workload note. * Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. * Default-zero rates are explicit assumptions, not evidence that the deployed service or resource is free. * Provisioned-unit minimums use an editable approximate allocation floor. Modeled target service cost is max(usage/allocation, minimum), with billable network, durable staging and dependent sinks added separately; actual RU/TU sizing, scaling increments and throughput remain unqualified. * Durable staging is priced once: provisioned targets include adapter state in their selected allocation; native targets add a separate staging allocation when required. Adapter hosting is a separate allocation and must exclude resources already priced as staging, workers or target capacity. * A declared adapter mapping and cost model do not establish implementation availability or service equivalence. * This is a declared directed mapping; a planning estimate does not establish deployment or request-serving support. * Origin unit prices are cited provider list prices; capacity coefficients and translation assumptions are editable authored planning guesses, not measured capacity or a provider quote. * Origin tariffs are the cited Claude Sonnet 4.6 global-endpoint list prices; target tariffs name no asserted current model SKU. Select model-specific rates before using the estimate. * Input and output meters and token expansion remain independent. A cheaper model is not automatically equivalent in capability or quality. * All target rates are editable authored planning guesses, not verified current list prices, customer discounts or benchmark evidence. * Dedicated accelerator/model-fit target: the Bedrock family returns this direction as Unpriced because the serving GPU or model deployment is not selected, so these generic rates are not shown as list prices. Scaleway Managed Inference lists EUR 0.93/hour for an L4-1-24G node ([https://www.scaleway.com/en/pricing/model-as-a-service/](https://www.scaleway.com/en/pricing/model-as-a-service/)); the Azure RAI policy has no meter of its own, its tokens bill on the Azure OpenAI deployment and knowledge bases on Azure AI Search (Basic USD 0.101/hour); the Tensor9 in-cluster model service needs GPU node pricing. Read 2026-10-22. * Row-store throughput (capacityOpsPerVcpuSecond, default 1500) is measured-by-tensor9 as a lower bound: the DynamoDB adapter on Cloud SQL db-custom-2-7680 (2 vCPU, PostgreSQL 16, us-west1) served 2,974 ops/s of YCSB-A at 16 threads without recorded saturation (src/main/kotlin/t9/prime/equiv/generated/eval/CloudsqlPostgresqlYcsbAT16GoYcsbV1\_0\_0.kt). Published CPU-bound pgbench reaches 3,101 read-write transactions and 20,289 point reads per vCPU-second (AWS z1d.2xlarge, PostgreSQL 14.5), so actual capacity is likely higher. Reads and writes share one coefficient: at the same adapter version, read-only YCSB-C was not faster than 50%-update YCSB-A. * Log-broker throughput (capacityStreamMiBPerVcpuSecond 20 at 0.5 utilization, 4 GiB RAM per vCPU) is provider-documented by Google Managed Service for Apache Kafka sizing ([https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size](https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size)): write-equivalent bytes are replicated writes plus a quarter of reads. Producer batches below 10 KB lower per-CPU throughput; that penalty is not modeled. * Disk utilization 0.8 is provider-documented headroom below Cloud SQL automatic storage increase, which triggers near 91-95% full ([https://docs.cloud.google.com/sql/docs/postgres/instance-settings](https://docs.cloud.google.com/sql/docs/postgres/instance-settings)). The 20 GiB reserve exceeds PostgreSQL's 1 GB default max\_wal\_size; the remainder of that reserve is unmeasured. * CPU tiers, 4 GiB per vCPU, at most 6.5 GiB per vCPU, 96 vCPU and 65,536 GiB disk follow Cloud SQL Enterprise custom machine limits (provider-documented, same page); 624 GiB is 96 × 6.5. Targets billed per node, unit or throughput unit are sized by their backend instead. * No public evidence: peak/mean work (2) and the RAM working-set fraction (0.1). Proposed measurements: the p99 one-minute rate over the monthly mean from customer CloudWatch metrics; the buffer hit ratio while stepping RAM under a fixed YCSB-B load. * Stored GiB already includes the origin family's physical footprint assumptions; no second storage amplification is applied. * The operations/vCPU coefficient supplies the compute requirement alongside explicit resource-hour minima. * Copies and allocation fractions change cost only; they do not increase modeled primary capacity. * AWS Price List API, AmazonBedrockFoundationModels offer: [https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json) Cost scope: Modeled target native service and explicitly modeled state allocation; hosting and Tensor9 fees only when supplied. Excluded costs and workload features: * Prompt caching, batch discounts, context tiers, grounding, tools, agents, embeddings, search, guardrails and dedicated inference hardware. * Taxes, discounts, credits and free-tier allowances unless explicitly reflected in edited rates. * Unmodeled dependent services, task execution and operations outside the selected workload family. List prices read: 2026-10-22. Model: `plan-a6f9604e9c2919f36a8bcff1239ddada7e8d8663` / `2026-10-22.planning.2`. Curve: `workload-2`. Sources: [AWS Price List API, AmazonBedrockFoundationModels offer](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json). ## Bedrock model inference → Vertex AI - Claude Origin API: `aws::1.0.0::bedrock`. Target backend: `google::1.0.0::vertex::claude`. ### Bedrock model inference · Input tokens per second Estimated monthly service charges by Input tokens per second. Adapter hosting and Tensor9 fees are excluded. Source: us-east-1; target: us-east5; currency: USD. ![Bedrock model inference · Input tokens per second: monthly service charges by Input tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/401d6c4811380b05dcd2c3e03f729da800eac78acb3e17fd96fbcc7f1b162191-light.svg) ![Bedrock model inference · Input tokens per second: monthly service charges by Input tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/6685ff59358034502b02c1cb8f941b58a7e5c269563d3c01271974bf4a0f9715-dark.svg) Comparison pills describe all sampled scale points; card amounts show the named workload. Costs The Same means within 20% at every point. Costs Depend includes mixed results and unpriced samples. Exact sampled model outputs in USD/month. Difference means target minus origin; a positive difference means the target costs more. | Input tokens per second (tokens/s) | Bedrock (LLM inference) | Target scenario | Vertex AI - Claude | Difference | | -: | -: | - | -: | -: | | 0 | 375 | Vertex AI - Claude | 412.5 | 37.5 | | 10 | 453.84 | Vertex AI - Claude | 499.224 | 45.384 | | 100 | 1163.4 | Vertex AI - Claude | 1279.74 | 116.34 | | 1000 | 8259 | Vertex AI - Claude | 9084.9 | 825.9 | | 10000 | 79215 | Vertex AI - Claude | 87136.5 | 7921.5 | | 100000 | 788775 | Vertex AI - Claude | 867652.5 | 78877.5 | **Curve assumptions** | Target scenario | Assumption | Value | Unit | | - | - | -: | - | | Vertex AI - Claude | Origin list price · Input tokens | 0.000003 | USD/Tokens | | Vertex AI - Claude | Origin list price · Output tokens | 0.000015 | USD/Tokens | | Vertex AI - Claude | Target / origin input tokens | 1 | multiplier | | Vertex AI - Claude | Target / origin output tokens | 1 | multiplier | | Vertex AI - Claude | Target price guess · read | 0 | USD/read | | Vertex AI - Claude | Target price guess · write | 0 | USD/write | | Vertex AI - Claude | Target price guess · provider operation | 0 | USD/provider operation | | Vertex AI - Claude | Target price guess · GiB-month | 0 | USD/GiB-month | | Vertex AI - Claude | Target price guess · published GiB | 0 | USD/published GiB | | Vertex AI - Claude | Target price guess · delivered GiB | 0 | USD/delivered GiB | | Vertex AI - Claude | Target price guess · network GiB | 0 | USD/network GiB | | Vertex AI - Claude | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | Vertex AI - Claude | Target price guess · vCPU-hour | 0 | USD/vCPU-hour | | Vertex AI - Claude | Target price guess · GiB-hour | 0 | USD/GiB-hour | | Vertex AI - Claude | Target price guess · resource-hour | 0 | USD/resource-hour | | Vertex AI - Claude | Target price guess · resource-month | 0 | USD/resource-month | | Vertex AI - Claude | Target list price · input token | 0.0000033 | USD/input token | | Vertex AI - Claude | Target list price · output token | 0.0000165 | USD/output token | | Vertex AI - Claude | Target monthly base charge | 0 | USD/month | | Vertex AI - Claude | Approximate target service minimum | 0 | USD/month | | Vertex AI - Claude | Minimum primary CPU | 2 | vCPU | | Vertex AI - Claude | Minimum primary RAM | 8 | GiB | | Vertex AI - Claude | Minimum primary disk | 100 | GiB | | Vertex AI - Claude | Additional full-price standby copy | 0 | 0 or 1 | | Vertex AI - Claude | Allocated share of provisioned cost | 1 | fraction | | Vertex AI - Claude | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | Vertex AI - Claude | Planning peak / mean work | 2 | multiplier | | Vertex AI - Claude | Planning primary CPU utilization | 0.7 | fraction | | Vertex AI - Claude | Resident state reserved in RAM | 0.1 | fraction | | Vertex AI - Claude | Planning disk utilization | 0.8 | fraction | | Vertex AI - Claude | Maintenance and log disk reserve | 20 | GiB | | Vertex AI - Claude | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | Vertex AI - Claude | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | Vertex AI - Claude | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | Vertex AI - Claude | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | Vertex AI - Claude | Dependent object sink write price guess | 0.000005 | USD/write | * Only this native meter varies; all other source usage and translation assumptions retain the displayed defaults. Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. Other native meters retain their explicitly displayed default quantities. * Planning estimate · editable authored rates and sizing assumptions dated 2026-10-22; not verified provider prices. us-east-1 → us-east5, USD/month. * Source prices are reference charges, not an invoice. Target lines cover selected service usage/configuration only; adapter hosting and Tensor9 fees are excluded. * Explicit tariff samples are not performance measurements or capacity qualification. No interpolation or break-even guarantee is implied. * Variant assumptions show the first axis sample; any axis-dependent configuration is described by this curve's workload note. * Rates are cited provider list prices, editable for discounts, commitments or another region. Workload translation and capacity coefficients remain editable planning assumptions, not benchmark measurements or a commercial quote. * Default-zero rates are explicit assumptions, not evidence that the deployed service or resource is free. * Provisioned-unit minimums use an editable approximate allocation floor. Modeled target service cost is max(usage/allocation, minimum), with billable network, durable staging and dependent sinks added separately; actual RU/TU sizing, scaling increments and throughput remain unqualified. * Durable staging is priced once: provisioned targets include adapter state in their selected allocation; native targets add a separate staging allocation when required. Adapter hosting is a separate allocation and must exclude resources already priced as staging, workers or target capacity. * A declared adapter mapping and cost model do not establish implementation availability or service equivalence. * This is a declared directed mapping; a planning estimate does not establish deployment or request-serving support. * Origin unit prices are cited provider list prices; capacity coefficients and translation assumptions are editable authored planning guesses, not measured capacity or a provider quote. * Origin tariffs are the cited Claude Sonnet 4.6 global-endpoint list prices; target tariffs name no asserted current model SKU. Select model-specific rates before using the estimate. * Input and output meters and token expansion remain independent. A cheaper model is not automatically equivalent in capability or quality. * Input and output token tariffs remain distinct and establish no equivalent output quality or tokenizer. Claude Sonnet 4.6 on the us-east5 regional endpoint the compiler stamps; regional and multi-region endpoints cost 10% more over the global endpoint (3.00/15.00). * AWS Price List API, AmazonBedrockFoundationModels offer: [https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json) * Vertex AI generative AI pricing: [https://cloud.google.com/vertex-ai/generative-ai/pricing](https://cloud.google.com/vertex-ai/generative-ai/pricing) Cost scope: Modeled target native service and explicitly modeled state allocation; hosting and Tensor9 fees only when supplied. Excluded costs and workload features: * Prompt caching, batch discounts, context tiers, grounding, tools, agents, embeddings, search, guardrails and dedicated inference hardware. * Taxes, discounts, credits and free-tier allowances unless explicitly reflected in edited rates. * Unmodeled dependent services, task execution and operations outside the selected workload family. List prices read: 2026-10-22. Model: `plan-48c312b989270b8c0eaf604ff9da8bb4641b8a7b` / `2026-10-22.planning.2`. Curve: `workload-1`. Sources: [AWS Price List API, AmazonBedrockFoundationModels offer](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json), [Vertex AI generative AI pricing](https://cloud.google.com/vertex-ai/generative-ai/pricing). ### Bedrock model inference · Output tokens per second Estimated monthly service charges by Output tokens per second. Adapter hosting and Tensor9 fees are excluded. Source: us-east-1; target: us-east5; currency: USD. ![Bedrock model inference · Output tokens per second: monthly service charges by Output tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/31513a053ac7bf916cc363bd336a223e1820ceefe2d56a36350908fc60f7448a-light.svg) ![Bedrock model inference · Output tokens per second: monthly service charges by Output tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/625454d9854b29bb2a76db1ec17e770c050644601dd8aec49fba06cc0731265b-dark.svg) Comparison pills describe all sampled scale points; card amounts show the named workload. Costs The Same means within 20% at every point. Costs Depend includes mixed results and unpriced samples. Exact sampled model outputs in USD/month. Difference means target minus origin; a positive difference means the target costs more. | Output tokens per second (tokens/s) | Bedrock (LLM inference) | Target scenario | Vertex AI - Claude | Difference | | -: | -: | - | -: | -: | | 0 | 300 | Vertex AI - Claude | 330 | 30 | | 10 | 694.2 | Vertex AI - Claude | 763.62 | 69.42 | | 100 | 4242 | Vertex AI - Claude | 4666.2 | 424.2 | | 1000 | 39720 | Vertex AI - Claude | 43692 | 3972 | | 10000 | 394500 | Vertex AI - Claude | 433950 | 39450 | | 100000 | 3942300 | Vertex AI - Claude | 4336530 | 394230 | **Curve assumptions** | Target scenario | Assumption | Value | Unit | | - | - | -: | - | | Vertex AI - Claude | Origin list price · Input tokens | 0.000003 | USD/Tokens | | Vertex AI - Claude | Origin list price · Output tokens | 0.000015 | USD/Tokens | | Vertex AI - Claude | Target / origin input tokens | 1 | multiplier | | Vertex AI - Claude | Target / origin output tokens | 1 | multiplier | | Vertex AI - Claude | Target price guess · read | 0 | USD/read | | Vertex AI - Claude | Target price guess · write | 0 | USD/write | | Vertex AI - Claude | Target price guess · provider operation | 0 | USD/provider operation | | Vertex AI - Claude | Target price guess · GiB-month | 0 | USD/GiB-month | | Vertex AI - Claude | Target price guess · published GiB | 0 | USD/published GiB | | Vertex AI - Claude | Target price guess · delivered GiB | 0 | USD/delivered GiB | | Vertex AI - Claude | Target price guess · network GiB | 0 | USD/network GiB | | Vertex AI - Claude | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | Vertex AI - Claude | Target price guess · vCPU-hour | 0 | USD/vCPU-hour | | Vertex AI - Claude | Target price guess · GiB-hour | 0 | USD/GiB-hour | | Vertex AI - Claude | Target price guess · resource-hour | 0 | USD/resource-hour | | Vertex AI - Claude | Target price guess · resource-month | 0 | USD/resource-month | | Vertex AI - Claude | Target list price · input token | 0.0000033 | USD/input token | | Vertex AI - Claude | Target list price · output token | 0.0000165 | USD/output token | | Vertex AI - Claude | Target monthly base charge | 0 | USD/month | | Vertex AI - Claude | Approximate target service minimum | 0 | USD/month | | Vertex AI - Claude | Minimum primary CPU | 2 | vCPU | | Vertex AI - Claude | Minimum primary RAM | 8 | GiB | | Vertex AI - Claude | Minimum primary disk | 100 | GiB | | Vertex AI - Claude | Additional full-price standby copy | 0 | 0 or 1 | | Vertex AI - Claude | Allocated share of provisioned cost | 1 | fraction | | Vertex AI - Claude | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | Vertex AI - Claude | Planning peak / mean work | 2 | multiplier | | Vertex AI - Claude | Planning primary CPU utilization | 0.7 | fraction | | Vertex AI - Claude | Resident state reserved in RAM | 0.1 | fraction | | Vertex AI - Claude | Planning disk utilization | 0.8 | fraction | | Vertex AI - Claude | Maintenance and log disk reserve | 20 | GiB | | Vertex AI - Claude | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | Vertex AI - Claude | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | Vertex AI - Claude | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | Vertex AI - Claude | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | Vertex AI - Claude | Dependent object sink write price guess | 0.000005 | USD/write | * Only this native meter varies; all other source usage and translation assumptions retain the displayed defaults. Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. Other native meters retain their explicitly displayed default quantities. * Planning estimate · editable authored rates and sizing assumptions dated 2026-10-22; not verified provider prices. us-east-1 → us-east5, USD/month. * Source prices are reference charges, not an invoice. Target lines cover selected service usage/configuration only; adapter hosting and Tensor9 fees are excluded. * Explicit tariff samples are not performance measurements or capacity qualification. No interpolation or break-even guarantee is implied. * Variant assumptions show the first axis sample; any axis-dependent configuration is described by this curve's workload note. * Rates are cited provider list prices, editable for discounts, commitments or another region. Workload translation and capacity coefficients remain editable planning assumptions, not benchmark measurements or a commercial quote. * Default-zero rates are explicit assumptions, not evidence that the deployed service or resource is free. * Provisioned-unit minimums use an editable approximate allocation floor. Modeled target service cost is max(usage/allocation, minimum), with billable network, durable staging and dependent sinks added separately; actual RU/TU sizing, scaling increments and throughput remain unqualified. * Durable staging is priced once: provisioned targets include adapter state in their selected allocation; native targets add a separate staging allocation when required. Adapter hosting is a separate allocation and must exclude resources already priced as staging, workers or target capacity. * A declared adapter mapping and cost model do not establish implementation availability or service equivalence. * This is a declared directed mapping; a planning estimate does not establish deployment or request-serving support. * Origin unit prices are cited provider list prices; capacity coefficients and translation assumptions are editable authored planning guesses, not measured capacity or a provider quote. * Origin tariffs are the cited Claude Sonnet 4.6 global-endpoint list prices; target tariffs name no asserted current model SKU. Select model-specific rates before using the estimate. * Input and output meters and token expansion remain independent. A cheaper model is not automatically equivalent in capability or quality. * Input and output token tariffs remain distinct and establish no equivalent output quality or tokenizer. Claude Sonnet 4.6 on the us-east5 regional endpoint the compiler stamps; regional and multi-region endpoints cost 10% more over the global endpoint (3.00/15.00). * AWS Price List API, AmazonBedrockFoundationModels offer: [https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json) * Vertex AI generative AI pricing: [https://cloud.google.com/vertex-ai/generative-ai/pricing](https://cloud.google.com/vertex-ai/generative-ai/pricing) Cost scope: Modeled target native service and explicitly modeled state allocation; hosting and Tensor9 fees only when supplied. Excluded costs and workload features: * Prompt caching, batch discounts, context tiers, grounding, tools, agents, embeddings, search, guardrails and dedicated inference hardware. * Taxes, discounts, credits and free-tier allowances unless explicitly reflected in edited rates. * Unmodeled dependent services, task execution and operations outside the selected workload family. List prices read: 2026-10-22. Model: `plan-48c312b989270b8c0eaf604ff9da8bb4641b8a7b` / `2026-10-22.planning.2`. Curve: `workload-2`. Sources: [AWS Price List API, AmazonBedrockFoundationModels offer](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json), [Vertex AI generative AI pricing](https://cloud.google.com/vertex-ai/generative-ai/pricing). ## Bedrock model inference → Vertex AI - Gemini Origin API: `aws::1.0.0::bedrock`. Target backend: `google::1.0.0::vertex::gemini`. ### Bedrock model inference · Input tokens per second Estimated monthly service charges by Input tokens per second. Adapter hosting and Tensor9 fees are excluded. Source: us-east-1; target: us-central1; currency: USD. ![Bedrock model inference · Input tokens per second: monthly service charges by Input tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/629afc5427a41e96349be140136f3a7413df5f7feb87e212c89851c347ed5a0c-light.svg) ![Bedrock model inference · Input tokens per second: monthly service charges by Input tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/fe430560dec5e47d1a9da7cc98e959648b542cb998ec1be494acebe118626c65-dark.svg) Comparison pills describe all sampled scale points; card amounts show the named workload. Costs The Same means within 20% at every point. Costs Depend includes mixed results and unpriced samples. Exact sampled model outputs in USD/month. Difference means target minus origin; a positive difference means the target costs more. | Input tokens per second (tokens/s) | Bedrock (LLM inference) | Target scenario | Vertex AI - Gemini | Difference | | -: | -: | - | -: | -: | | 0 | 375 | Vertex AI - Gemini | 62.5 | -312.5 | | 10 | 453.84 | Vertex AI - Gemini | 70.384 | -383.456 | | 100 | 1163.4 | Vertex AI - Gemini | 141.34 | -1022.06 | | 1000 | 8259 | Vertex AI - Gemini | 850.9 | -7408.1 | | 10000 | 79215 | Vertex AI - Gemini | 7946.5 | -71268.5 | | 100000 | 788775 | Vertex AI - Gemini | 78902.5 | -709872.5 | **Curve assumptions** | Target scenario | Assumption | Value | Unit | | - | - | -: | - | | Vertex AI - Gemini | Origin list price · Input tokens | 0.000003 | USD/Tokens | | Vertex AI - Gemini | Origin list price · Output tokens | 0.000015 | USD/Tokens | | Vertex AI - Gemini | Target / origin input tokens | 1 | multiplier | | Vertex AI - Gemini | Target / origin output tokens | 1 | multiplier | | Vertex AI - Gemini | Target price guess · read | 0 | USD/read | | Vertex AI - Gemini | Target price guess · write | 0 | USD/write | | Vertex AI - Gemini | Target price guess · provider operation | 0 | USD/provider operation | | Vertex AI - Gemini | Target price guess · GiB-month | 0 | USD/GiB-month | | Vertex AI - Gemini | Target price guess · published GiB | 0 | USD/published GiB | | Vertex AI - Gemini | Target price guess · delivered GiB | 0 | USD/delivered GiB | | Vertex AI - Gemini | Target price guess · network GiB | 0 | USD/network GiB | | Vertex AI - Gemini | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | Vertex AI - Gemini | Target price guess · vCPU-hour | 0 | USD/vCPU-hour | | Vertex AI - Gemini | Target price guess · GiB-hour | 0 | USD/GiB-hour | | Vertex AI - Gemini | Target price guess · resource-hour | 0 | USD/resource-hour | | Vertex AI - Gemini | Target price guess · resource-month | 0 | USD/resource-month | | Vertex AI - Gemini | Target list price · input token | 0.0000003 | USD/input token | | Vertex AI - Gemini | Target list price · output token | 0.0000025 | USD/output token | | Vertex AI - Gemini | Target monthly base charge | 0 | USD/month | | Vertex AI - Gemini | Approximate target service minimum | 0 | USD/month | | Vertex AI - Gemini | Minimum primary CPU | 2 | vCPU | | Vertex AI - Gemini | Minimum primary RAM | 8 | GiB | | Vertex AI - Gemini | Minimum primary disk | 100 | GiB | | Vertex AI - Gemini | Additional full-price standby copy | 0 | 0 or 1 | | Vertex AI - Gemini | Allocated share of provisioned cost | 1 | fraction | | Vertex AI - Gemini | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | Vertex AI - Gemini | Planning peak / mean work | 2 | multiplier | | Vertex AI - Gemini | Planning primary CPU utilization | 0.7 | fraction | | Vertex AI - Gemini | Resident state reserved in RAM | 0.1 | fraction | | Vertex AI - Gemini | Planning disk utilization | 0.8 | fraction | | Vertex AI - Gemini | Maintenance and log disk reserve | 20 | GiB | | Vertex AI - Gemini | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | Vertex AI - Gemini | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | Vertex AI - Gemini | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | Vertex AI - Gemini | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | Vertex AI - Gemini | Dependent object sink write price guess | 0.000005 | USD/write | * Only this native meter varies; all other source usage and translation assumptions retain the displayed defaults. Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. Other native meters retain their explicitly displayed default quantities. * Planning estimate · editable authored rates and sizing assumptions dated 2026-10-22; not verified provider prices. us-east-1 → us-central1, USD/month. * Source prices are reference charges, not an invoice. Target lines cover selected service usage/configuration only; adapter hosting and Tensor9 fees are excluded. * Explicit tariff samples are not performance measurements or capacity qualification. No interpolation or break-even guarantee is implied. * Variant assumptions show the first axis sample; any axis-dependent configuration is described by this curve's workload note. * Rates are cited provider list prices, editable for discounts, commitments or another region. Workload translation and capacity coefficients remain editable planning assumptions, not benchmark measurements or a commercial quote. * Default-zero rates are explicit assumptions, not evidence that the deployed service or resource is free. * Provisioned-unit minimums use an editable approximate allocation floor. Modeled target service cost is max(usage/allocation, minimum), with billable network, durable staging and dependent sinks added separately; actual RU/TU sizing, scaling increments and throughput remain unqualified. * Durable staging is priced once: provisioned targets include adapter state in their selected allocation; native targets add a separate staging allocation when required. Adapter hosting is a separate allocation and must exclude resources already priced as staging, workers or target capacity. * A declared adapter mapping and cost model do not establish implementation availability or service equivalence. * This is a declared directed mapping; a planning estimate does not establish deployment or request-serving support. * Origin unit prices are cited provider list prices; capacity coefficients and translation assumptions are editable authored planning guesses, not measured capacity or a provider quote. * Origin tariffs are the cited Claude Sonnet 4.6 global-endpoint list prices; target tariffs name no asserted current model SKU. Select model-specific rates before using the estimate. * Input and output meters and token expansion remain independent. A cheaper model is not automatically equivalent in capability or quality. * Input and output token tariffs remain distinct and establish no equivalent output quality or tokenizer. gemini-2.5-flash, the model the compiler stamps when none is supplied. Output includes reasoning tokens; audio input lists 1.00 per 1M. * AWS Price List API, AmazonBedrockFoundationModels offer: [https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json) * Vertex AI generative AI pricing: [https://cloud.google.com/vertex-ai/generative-ai/pricing](https://cloud.google.com/vertex-ai/generative-ai/pricing) Cost scope: Modeled target native service and explicitly modeled state allocation; hosting and Tensor9 fees only when supplied. Excluded costs and workload features: * Prompt caching, batch discounts, context tiers, grounding, tools, agents, embeddings, search, guardrails and dedicated inference hardware. * Taxes, discounts, credits and free-tier allowances unless explicitly reflected in edited rates. * Unmodeled dependent services, task execution and operations outside the selected workload family. List prices read: 2026-10-22. Model: `plan-4c72ad085195d9377822f40afb60d644ed875746` / `2026-10-22.planning.2`. Curve: `workload-1`. Sources: [AWS Price List API, AmazonBedrockFoundationModels offer](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json), [Vertex AI generative AI pricing](https://cloud.google.com/vertex-ai/generative-ai/pricing). ### Bedrock model inference · Output tokens per second Estimated monthly service charges by Output tokens per second. Adapter hosting and Tensor9 fees are excluded. Source: us-east-1; target: us-central1; currency: USD. ![Bedrock model inference · Output tokens per second: monthly service charges by Output tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/af003f4c49c7c9c71d4d9a880e4af7b055d32d41a6b804babcca3c2e70abd922-light.svg) ![Bedrock model inference · Output tokens per second: monthly service charges by Output tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/3025224399ded5a11c1c51db662acf71167bad640857b3860e68cbf215b796dc-dark.svg) Comparison pills describe all sampled scale points; card amounts show the named workload. Costs The Same means within 20% at every point. Costs Depend includes mixed results and unpriced samples. Exact sampled model outputs in USD/month. Difference means target minus origin; a positive difference means the target costs more. | Output tokens per second (tokens/s) | Bedrock (LLM inference) | Target scenario | Vertex AI - Gemini | Difference | | -: | -: | - | -: | -: | | 0 | 300 | Vertex AI - Gemini | 30 | -270 | | 10 | 694.2 | Vertex AI - Gemini | 95.7 | -598.5 | | 100 | 4242 | Vertex AI - Gemini | 687 | -3555 | | 1000 | 39720 | Vertex AI - Gemini | 6600 | -33120 | | 10000 | 394500 | Vertex AI - Gemini | 65730 | -328770 | | 100000 | 3942300 | Vertex AI - Gemini | 657030 | -3285270 | **Curve assumptions** | Target scenario | Assumption | Value | Unit | | - | - | -: | - | | Vertex AI - Gemini | Origin list price · Input tokens | 0.000003 | USD/Tokens | | Vertex AI - Gemini | Origin list price · Output tokens | 0.000015 | USD/Tokens | | Vertex AI - Gemini | Target / origin input tokens | 1 | multiplier | | Vertex AI - Gemini | Target / origin output tokens | 1 | multiplier | | Vertex AI - Gemini | Target price guess · read | 0 | USD/read | | Vertex AI - Gemini | Target price guess · write | 0 | USD/write | | Vertex AI - Gemini | Target price guess · provider operation | 0 | USD/provider operation | | Vertex AI - Gemini | Target price guess · GiB-month | 0 | USD/GiB-month | | Vertex AI - Gemini | Target price guess · published GiB | 0 | USD/published GiB | | Vertex AI - Gemini | Target price guess · delivered GiB | 0 | USD/delivered GiB | | Vertex AI - Gemini | Target price guess · network GiB | 0 | USD/network GiB | | Vertex AI - Gemini | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | Vertex AI - Gemini | Target price guess · vCPU-hour | 0 | USD/vCPU-hour | | Vertex AI - Gemini | Target price guess · GiB-hour | 0 | USD/GiB-hour | | Vertex AI - Gemini | Target price guess · resource-hour | 0 | USD/resource-hour | | Vertex AI - Gemini | Target price guess · resource-month | 0 | USD/resource-month | | Vertex AI - Gemini | Target list price · input token | 0.0000003 | USD/input token | | Vertex AI - Gemini | Target list price · output token | 0.0000025 | USD/output token | | Vertex AI - Gemini | Target monthly base charge | 0 | USD/month | | Vertex AI - Gemini | Approximate target service minimum | 0 | USD/month | | Vertex AI - Gemini | Minimum primary CPU | 2 | vCPU | | Vertex AI - Gemini | Minimum primary RAM | 8 | GiB | | Vertex AI - Gemini | Minimum primary disk | 100 | GiB | | Vertex AI - Gemini | Additional full-price standby copy | 0 | 0 or 1 | | Vertex AI - Gemini | Allocated share of provisioned cost | 1 | fraction | | Vertex AI - Gemini | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | Vertex AI - Gemini | Planning peak / mean work | 2 | multiplier | | Vertex AI - Gemini | Planning primary CPU utilization | 0.7 | fraction | | Vertex AI - Gemini | Resident state reserved in RAM | 0.1 | fraction | | Vertex AI - Gemini | Planning disk utilization | 0.8 | fraction | | Vertex AI - Gemini | Maintenance and log disk reserve | 20 | GiB | | Vertex AI - Gemini | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | Vertex AI - Gemini | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | Vertex AI - Gemini | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | Vertex AI - Gemini | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | Vertex AI - Gemini | Dependent object sink write price guess | 0.000005 | USD/write | * Only this native meter varies; all other source usage and translation assumptions retain the displayed defaults. Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. Other native meters retain their explicitly displayed default quantities. * Planning estimate · editable authored rates and sizing assumptions dated 2026-10-22; not verified provider prices. us-east-1 → us-central1, USD/month. * Source prices are reference charges, not an invoice. Target lines cover selected service usage/configuration only; adapter hosting and Tensor9 fees are excluded. * Explicit tariff samples are not performance measurements or capacity qualification. No interpolation or break-even guarantee is implied. * Variant assumptions show the first axis sample; any axis-dependent configuration is described by this curve's workload note. * Rates are cited provider list prices, editable for discounts, commitments or another region. Workload translation and capacity coefficients remain editable planning assumptions, not benchmark measurements or a commercial quote. * Default-zero rates are explicit assumptions, not evidence that the deployed service or resource is free. * Provisioned-unit minimums use an editable approximate allocation floor. Modeled target service cost is max(usage/allocation, minimum), with billable network, durable staging and dependent sinks added separately; actual RU/TU sizing, scaling increments and throughput remain unqualified. * Durable staging is priced once: provisioned targets include adapter state in their selected allocation; native targets add a separate staging allocation when required. Adapter hosting is a separate allocation and must exclude resources already priced as staging, workers or target capacity. * A declared adapter mapping and cost model do not establish implementation availability or service equivalence. * This is a declared directed mapping; a planning estimate does not establish deployment or request-serving support. * Origin unit prices are cited provider list prices; capacity coefficients and translation assumptions are editable authored planning guesses, not measured capacity or a provider quote. * Origin tariffs are the cited Claude Sonnet 4.6 global-endpoint list prices; target tariffs name no asserted current model SKU. Select model-specific rates before using the estimate. * Input and output meters and token expansion remain independent. A cheaper model is not automatically equivalent in capability or quality. * Input and output token tariffs remain distinct and establish no equivalent output quality or tokenizer. gemini-2.5-flash, the model the compiler stamps when none is supplied. Output includes reasoning tokens; audio input lists 1.00 per 1M. * AWS Price List API, AmazonBedrockFoundationModels offer: [https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json) * Vertex AI generative AI pricing: [https://cloud.google.com/vertex-ai/generative-ai/pricing](https://cloud.google.com/vertex-ai/generative-ai/pricing) Cost scope: Modeled target native service and explicitly modeled state allocation; hosting and Tensor9 fees only when supplied. Excluded costs and workload features: * Prompt caching, batch discounts, context tiers, grounding, tools, agents, embeddings, search, guardrails and dedicated inference hardware. * Taxes, discounts, credits and free-tier allowances unless explicitly reflected in edited rates. * Unmodeled dependent services, task execution and operations outside the selected workload family. List prices read: 2026-10-22. Model: `plan-4c72ad085195d9377822f40afb60d644ed875746` / `2026-10-22.planning.2`. Curve: `workload-2`. Sources: [AWS Price List API, AmazonBedrockFoundationModels offer](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json), [Vertex AI generative AI pricing](https://cloud.google.com/vertex-ai/generative-ai/pricing). ## Bedrock model inference → Vertex AI - Llama Origin API: `aws::1.0.0::bedrock`. Target backend: `google::1.0.0::vertex::llama`. ### Bedrock model inference · Input tokens per second Estimated monthly service charges by Input tokens per second. Adapter hosting and Tensor9 fees are excluded. Source: us-east-1; target: us-central1; currency: USD. ![Bedrock model inference · Input tokens per second: monthly service charges by Input tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/66eb5ecae403c933b72f743191879a2d7e258e0db1fef406efb7d4b819ee8242-light.svg) ![Bedrock model inference · Input tokens per second: monthly service charges by Input tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/8576beaf72ebbac1ea38363e4aefa43740d0a0e56a81176c36f4ed9ef34d2268-dark.svg) Comparison pills describe all sampled scale points; card amounts show the named workload. Costs The Same means within 20% at every point. Costs Depend includes mixed results and unpriced samples. Exact sampled model outputs in USD/month. Difference means target minus origin; a positive difference means the target costs more. | Input tokens per second (tokens/s) | Bedrock (LLM inference) | Target scenario | Vertex AI - Llama | Difference | | -: | -: | - | -: | -: | | 0 | 375 | Vertex AI - Llama | 18 | -357 | | 10 | 453.84 | Vertex AI - Llama | 36.9216 | -416.9184 | | 100 | 1163.4 | Vertex AI - Llama | 207.216 | -956.184 | | 1000 | 8259 | Vertex AI - Llama | 1910.16 | -6348.84 | | 10000 | 79215 | Vertex AI - Llama | 18939.6 | -60275.4 | | 100000 | 788775 | Vertex AI - Llama | 189234 | -599541 | **Curve assumptions** | Target scenario | Assumption | Value | Unit | | - | - | -: | - | | Vertex AI - Llama | Origin list price · Input tokens | 0.000003 | USD/Tokens | | Vertex AI - Llama | Origin list price · Output tokens | 0.000015 | USD/Tokens | | Vertex AI - Llama | Target / origin input tokens | 1 | multiplier | | Vertex AI - Llama | Target / origin output tokens | 1 | multiplier | | Vertex AI - Llama | Target price guess · read | 0 | USD/read | | Vertex AI - Llama | Target price guess · write | 0 | USD/write | | Vertex AI - Llama | Target price guess · provider operation | 0 | USD/provider operation | | Vertex AI - Llama | Target price guess · GiB-month | 0 | USD/GiB-month | | Vertex AI - Llama | Target price guess · published GiB | 0 | USD/published GiB | | Vertex AI - Llama | Target price guess · delivered GiB | 0 | USD/delivered GiB | | Vertex AI - Llama | Target price guess · network GiB | 0 | USD/network GiB | | Vertex AI - Llama | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | Vertex AI - Llama | Target price guess · vCPU-hour | 0 | USD/vCPU-hour | | Vertex AI - Llama | Target price guess · GiB-hour | 0 | USD/GiB-hour | | Vertex AI - Llama | Target price guess · resource-hour | 0 | USD/resource-hour | | Vertex AI - Llama | Target price guess · resource-month | 0 | USD/resource-month | | Vertex AI - Llama | Target list price · input token | 0.00000072 | USD/input token | | Vertex AI - Llama | Target list price · output token | 0.00000072 | USD/output token | | Vertex AI - Llama | Target monthly base charge | 0 | USD/month | | Vertex AI - Llama | Approximate target service minimum | 0 | USD/month | | Vertex AI - Llama | Minimum primary CPU | 2 | vCPU | | Vertex AI - Llama | Minimum primary RAM | 8 | GiB | | Vertex AI - Llama | Minimum primary disk | 100 | GiB | | Vertex AI - Llama | Additional full-price standby copy | 0 | 0 or 1 | | Vertex AI - Llama | Allocated share of provisioned cost | 1 | fraction | | Vertex AI - Llama | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | Vertex AI - Llama | Planning peak / mean work | 2 | multiplier | | Vertex AI - Llama | Planning primary CPU utilization | 0.7 | fraction | | Vertex AI - Llama | Resident state reserved in RAM | 0.1 | fraction | | Vertex AI - Llama | Planning disk utilization | 0.8 | fraction | | Vertex AI - Llama | Maintenance and log disk reserve | 20 | GiB | | Vertex AI - Llama | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | Vertex AI - Llama | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | Vertex AI - Llama | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | Vertex AI - Llama | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | Vertex AI - Llama | Dependent object sink write price guess | 0.000005 | USD/write | * Only this native meter varies; all other source usage and translation assumptions retain the displayed defaults. Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. Other native meters retain their explicitly displayed default quantities. * Planning estimate · editable authored rates and sizing assumptions dated 2026-10-22; not verified provider prices. us-east-1 → us-central1, USD/month. * Source prices are reference charges, not an invoice. Target lines cover selected service usage/configuration only; adapter hosting and Tensor9 fees are excluded. * Explicit tariff samples are not performance measurements or capacity qualification. No interpolation or break-even guarantee is implied. * Variant assumptions show the first axis sample; any axis-dependent configuration is described by this curve's workload note. * Rates are cited provider list prices, editable for discounts, commitments or another region. Workload translation and capacity coefficients remain editable planning assumptions, not benchmark measurements or a commercial quote. * Default-zero rates are explicit assumptions, not evidence that the deployed service or resource is free. * Provisioned-unit minimums use an editable approximate allocation floor. Modeled target service cost is max(usage/allocation, minimum), with billable network, durable staging and dependent sinks added separately; actual RU/TU sizing, scaling increments and throughput remain unqualified. * Durable staging is priced once: provisioned targets include adapter state in their selected allocation; native targets add a separate staging allocation when required. Adapter hosting is a separate allocation and must exclude resources already priced as staging, workers or target capacity. * A declared adapter mapping and cost model do not establish implementation availability or service equivalence. * This is a declared directed mapping; a planning estimate does not establish deployment or request-serving support. * Origin unit prices are cited provider list prices; capacity coefficients and translation assumptions are editable authored planning guesses, not measured capacity or a provider quote. * Origin tariffs are the cited Claude Sonnet 4.6 global-endpoint list prices; target tariffs name no asserted current model SKU. Select model-specific rates before using the estimate. * Input and output meters and token expansion remain independent. A cheaper model is not automatically equivalent in capability or quality. * Input and output token tariffs remain distinct and establish no equivalent output quality or tokenizer. meta/llama-3.3-70b-instruct-maas, the model the compiler stamps. * AWS Price List API, AmazonBedrockFoundationModels offer: [https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json) * Vertex AI generative AI pricing: [https://cloud.google.com/vertex-ai/generative-ai/pricing](https://cloud.google.com/vertex-ai/generative-ai/pricing) Cost scope: Modeled target native service and explicitly modeled state allocation; hosting and Tensor9 fees only when supplied. Excluded costs and workload features: * Prompt caching, batch discounts, context tiers, grounding, tools, agents, embeddings, search, guardrails and dedicated inference hardware. * Taxes, discounts, credits and free-tier allowances unless explicitly reflected in edited rates. * Unmodeled dependent services, task execution and operations outside the selected workload family. List prices read: 2026-10-22. Model: `plan-5c6a290efb8ab2af09754f01caf66263075fec9b` / `2026-10-22.planning.2`. Curve: `workload-1`. Sources: [AWS Price List API, AmazonBedrockFoundationModels offer](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json), [Vertex AI generative AI pricing](https://cloud.google.com/vertex-ai/generative-ai/pricing). ### Bedrock model inference · Output tokens per second Estimated monthly service charges by Output tokens per second. Adapter hosting and Tensor9 fees are excluded. Source: us-east-1; target: us-central1; currency: USD. ![Bedrock model inference · Output tokens per second: monthly service charges by Output tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/fedf854a4d5bf0db9166ddb8f86a5765bc3724b424403c77c2b57d4f35e67312-light.svg) ![Bedrock model inference · Output tokens per second: monthly service charges by Output tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/44fa5d045d04284ba46da4206f9021fedb70f848a493ddae3567ee52d8e736e2-dark.svg) Comparison pills describe all sampled scale points; card amounts show the named workload. Costs The Same means within 20% at every point. Costs Depend includes mixed results and unpriced samples. Exact sampled model outputs in USD/month. Difference means target minus origin; a positive difference means the target costs more. | Output tokens per second (tokens/s) | Bedrock (LLM inference) | Target scenario | Vertex AI - Llama | Difference | | -: | -: | - | -: | -: | | 0 | 300 | Vertex AI - Llama | 72 | -228 | | 10 | 694.2 | Vertex AI - Llama | 90.9216 | -603.2784 | | 100 | 4242 | Vertex AI - Llama | 261.216 | -3980.784 | | 1000 | 39720 | Vertex AI - Llama | 1964.16 | -37755.84 | | 10000 | 394500 | Vertex AI - Llama | 18993.6 | -375506.4 | | 100000 | 3942300 | Vertex AI - Llama | 189288 | -3753012 | **Curve assumptions** | Target scenario | Assumption | Value | Unit | | - | - | -: | - | | Vertex AI - Llama | Origin list price · Input tokens | 0.000003 | USD/Tokens | | Vertex AI - Llama | Origin list price · Output tokens | 0.000015 | USD/Tokens | | Vertex AI - Llama | Target / origin input tokens | 1 | multiplier | | Vertex AI - Llama | Target / origin output tokens | 1 | multiplier | | Vertex AI - Llama | Target price guess · read | 0 | USD/read | | Vertex AI - Llama | Target price guess · write | 0 | USD/write | | Vertex AI - Llama | Target price guess · provider operation | 0 | USD/provider operation | | Vertex AI - Llama | Target price guess · GiB-month | 0 | USD/GiB-month | | Vertex AI - Llama | Target price guess · published GiB | 0 | USD/published GiB | | Vertex AI - Llama | Target price guess · delivered GiB | 0 | USD/delivered GiB | | Vertex AI - Llama | Target price guess · network GiB | 0 | USD/network GiB | | Vertex AI - Llama | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | Vertex AI - Llama | Target price guess · vCPU-hour | 0 | USD/vCPU-hour | | Vertex AI - Llama | Target price guess · GiB-hour | 0 | USD/GiB-hour | | Vertex AI - Llama | Target price guess · resource-hour | 0 | USD/resource-hour | | Vertex AI - Llama | Target price guess · resource-month | 0 | USD/resource-month | | Vertex AI - Llama | Target list price · input token | 0.00000072 | USD/input token | | Vertex AI - Llama | Target list price · output token | 0.00000072 | USD/output token | | Vertex AI - Llama | Target monthly base charge | 0 | USD/month | | Vertex AI - Llama | Approximate target service minimum | 0 | USD/month | | Vertex AI - Llama | Minimum primary CPU | 2 | vCPU | | Vertex AI - Llama | Minimum primary RAM | 8 | GiB | | Vertex AI - Llama | Minimum primary disk | 100 | GiB | | Vertex AI - Llama | Additional full-price standby copy | 0 | 0 or 1 | | Vertex AI - Llama | Allocated share of provisioned cost | 1 | fraction | | Vertex AI - Llama | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | Vertex AI - Llama | Planning peak / mean work | 2 | multiplier | | Vertex AI - Llama | Planning primary CPU utilization | 0.7 | fraction | | Vertex AI - Llama | Resident state reserved in RAM | 0.1 | fraction | | Vertex AI - Llama | Planning disk utilization | 0.8 | fraction | | Vertex AI - Llama | Maintenance and log disk reserve | 20 | GiB | | Vertex AI - Llama | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | Vertex AI - Llama | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | Vertex AI - Llama | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | Vertex AI - Llama | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | Vertex AI - Llama | Dependent object sink write price guess | 0.000005 | USD/write | * Only this native meter varies; all other source usage and translation assumptions retain the displayed defaults. Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. Other native meters retain their explicitly displayed default quantities. * Planning estimate · editable authored rates and sizing assumptions dated 2026-10-22; not verified provider prices. us-east-1 → us-central1, USD/month. * Source prices are reference charges, not an invoice. Target lines cover selected service usage/configuration only; adapter hosting and Tensor9 fees are excluded. * Explicit tariff samples are not performance measurements or capacity qualification. No interpolation or break-even guarantee is implied. * Variant assumptions show the first axis sample; any axis-dependent configuration is described by this curve's workload note. * Rates are cited provider list prices, editable for discounts, commitments or another region. Workload translation and capacity coefficients remain editable planning assumptions, not benchmark measurements or a commercial quote. * Default-zero rates are explicit assumptions, not evidence that the deployed service or resource is free. * Provisioned-unit minimums use an editable approximate allocation floor. Modeled target service cost is max(usage/allocation, minimum), with billable network, durable staging and dependent sinks added separately; actual RU/TU sizing, scaling increments and throughput remain unqualified. * Durable staging is priced once: provisioned targets include adapter state in their selected allocation; native targets add a separate staging allocation when required. Adapter hosting is a separate allocation and must exclude resources already priced as staging, workers or target capacity. * A declared adapter mapping and cost model do not establish implementation availability or service equivalence. * This is a declared directed mapping; a planning estimate does not establish deployment or request-serving support. * Origin unit prices are cited provider list prices; capacity coefficients and translation assumptions are editable authored planning guesses, not measured capacity or a provider quote. * Origin tariffs are the cited Claude Sonnet 4.6 global-endpoint list prices; target tariffs name no asserted current model SKU. Select model-specific rates before using the estimate. * Input and output meters and token expansion remain independent. A cheaper model is not automatically equivalent in capability or quality. * Input and output token tariffs remain distinct and establish no equivalent output quality or tokenizer. meta/llama-3.3-70b-instruct-maas, the model the compiler stamps. * AWS Price List API, AmazonBedrockFoundationModels offer: [https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json) * Vertex AI generative AI pricing: [https://cloud.google.com/vertex-ai/generative-ai/pricing](https://cloud.google.com/vertex-ai/generative-ai/pricing) Cost scope: Modeled target native service and explicitly modeled state allocation; hosting and Tensor9 fees only when supplied. Excluded costs and workload features: * Prompt caching, batch discounts, context tiers, grounding, tools, agents, embeddings, search, guardrails and dedicated inference hardware. * Taxes, discounts, credits and free-tier allowances unless explicitly reflected in edited rates. * Unmodeled dependent services, task execution and operations outside the selected workload family. List prices read: 2026-10-22. Model: `plan-5c6a290efb8ab2af09754f01caf66263075fec9b` / `2026-10-22.planning.2`. Curve: `workload-2`. Sources: [AWS Price List API, AmazonBedrockFoundationModels offer](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json), [Vertex AI generative AI pricing](https://cloud.google.com/vertex-ai/generative-ai/pricing). ## Bedrock model inference → vLLM / Ollama Origin API: `aws::1.0.0::bedrock`. Target backend: `tensor9::1.0.0::emu::aws/1.0.0/bedrock::incluster`. ### Bedrock model inference · Input tokens per second Estimated monthly service charges by Input tokens per second. Adapter hosting and Tensor9 fees are excluded. Source: us-east-1; target: Explicit target planning region; currency: USD. ![Bedrock model inference · Input tokens per second: monthly service charges by Input tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/402238675a8de0146212dbba0443fdba8ab3b1cb3f60548f2d4c1b477302ef37-light.svg) ![Bedrock model inference · Input tokens per second: monthly service charges by Input tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/957a22a1a6229d501d91ad29895d5f10c995fdb472e6bf462694d64ad1b91012-dark.svg) Comparison pills describe all sampled scale points; card amounts show the named workload. Costs The Same means within 20% at every point. Costs Depend includes mixed results and unpriced samples. Exact sampled model outputs in USD/month. Difference means target minus origin; a positive difference means the target costs more. Estimated allocation from planning assumptions; not benchmark calibrated. Lines connect evaluated samples; intermediate capacity and prices are not inferred. Unpriced means no target quote or cost difference is available within the sizing policy. | Input tokens per second (tokens/s) | Bedrock (LLM inference) | Target scenario | vLLM / Ollama | Difference | | -: | -: | - | -: | -: | | 0 | 375 | vLLM / Ollama (one copy) | Unpriced | Unavailable | | 0 | 375 | vLLM / Ollama (two copies) | Unpriced | Unavailable | | 10 | 453.84 | vLLM / Ollama (one copy) | Unpriced | Unavailable | | 10 | 453.84 | vLLM / Ollama (two copies) | Unpriced | Unavailable | | 100 | 1163.4 | vLLM / Ollama (one copy) | Unpriced | Unavailable | | 100 | 1163.4 | vLLM / Ollama (two copies) | Unpriced | Unavailable | | 1000 | 8259 | vLLM / Ollama (one copy) | Unpriced | Unavailable | | 1000 | 8259 | vLLM / Ollama (two copies) | Unpriced | Unavailable | | 10000 | 79215 | vLLM / Ollama (one copy) | Unpriced | Unavailable | | 10000 | 79215 | vLLM / Ollama (two copies) | Unpriced | Unavailable | | 100000 | 788775 | vLLM / Ollama (one copy) | Unpriced | Unavailable | | 100000 | 788775 | vLLM / Ollama (two copies) | Unpriced | Unavailable | **Estimated allocation from planning assumptions; not benchmark calibrated.** The sizing policy grows compute with translated work and peak headroom, memory with compute and working set, and disk with resident working state and reserve. Monthly usage is distinct from resident data; each workload's assumptions define the translation. HA changes pricing, not this policy's throughput estimate. Unpriced workloads exceed the policy's configuration range; no silent vertical cap or automatic sharding is assumed. | Sample / scenario | Basis | Mean / peak work/s | Required vCPU | Allocated vCPU | Memory GiB | SSD GiB | Resident GiB | Policy | Constraints | | - | - | -: | -: | -: | -: | -: | -: | - | - | | 0 tokens/s / vLLM / Ollama (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 0 tokens/s / vLLM / Ollama (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10 tokens/s / vLLM / Ollama (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10 tokens/s / vLLM / Ollama (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100 tokens/s / vLLM / Ollama (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100 tokens/s / vLLM / Ollama (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 1000 tokens/s / vLLM / Ollama (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 1000 tokens/s / vLLM / Ollama (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10000 tokens/s / vLLM / Ollama (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10000 tokens/s / vLLM / Ollama (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100000 tokens/s / vLLM / Ollama (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100000 tokens/s / vLLM / Ollama (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | * Row-store throughput (capacityOpsPerVcpuSecond, default 1500) is measured-by-tensor9 as a lower bound: the DynamoDB adapter on Cloud SQL db-custom-2-7680 (2 vCPU, PostgreSQL 16, us-west1) served 2,974 ops/s of YCSB-A at 16 threads without recorded saturation (src/main/kotlin/t9/prime/equiv/generated/eval/CloudsqlPostgresqlYcsbAT16GoYcsbV1\_0\_0.kt). Published CPU-bound pgbench reaches 3,101 read-write transactions and 20,289 point reads per vCPU-second (AWS z1d.2xlarge, PostgreSQL 14.5), so actual capacity is likely higher. Reads and writes share one coefficient: at the same adapter version, read-only YCSB-C was not faster than 50%-update YCSB-A. * Log-broker throughput (capacityStreamMiBPerVcpuSecond 20 at 0.5 utilization, 4 GiB RAM per vCPU) is provider-documented by Google Managed Service for Apache Kafka sizing ([https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size](https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size)): write-equivalent bytes are replicated writes plus a quarter of reads. Producer batches below 10 KB lower per-CPU throughput; that penalty is not modeled. * Disk utilization 0.8 is provider-documented headroom below Cloud SQL automatic storage increase, which triggers near 91-95% full ([https://docs.cloud.google.com/sql/docs/postgres/instance-settings](https://docs.cloud.google.com/sql/docs/postgres/instance-settings)). The 20 GiB reserve exceeds PostgreSQL's 1 GB default max\_wal\_size; the remainder of that reserve is unmeasured. * CPU tiers, 4 GiB per vCPU, at most 6.5 GiB per vCPU, 96 vCPU and 65,536 GiB disk follow Cloud SQL Enterprise custom machine limits (provider-documented, same page); 624 GiB is 96 × 6.5. Targets billed per node, unit or throughput unit are sized by their backend instead. * No public evidence: peak/mean work (2) and the RAM working-set fraction (0.1). Proposed measurements: the p99 one-minute rate over the monthly mean from customer CloudWatch metrics; the buffer hit ratio while stepping RAM under a fixed YCSB-B load. * Stored GiB already includes the origin family's physical footprint assumptions; no second storage amplification is applied. * The operations/vCPU coefficient supplies the compute requirement alongside explicit resource-hour minima. * Copies and allocation fractions change cost only; they do not increase modeled primary capacity. **Curve assumptions** | Target scenario | Assumption | Value | Unit | | - | - | -: | - | | vLLM / Ollama (one copy) | Origin list price · Input tokens | 0.000003 | USD/Tokens | | vLLM / Ollama (one copy) | Origin list price · Output tokens | 0.000015 | USD/Tokens | | vLLM / Ollama (one copy) | Target / origin input tokens | 1 | multiplier | | vLLM / Ollama (one copy) | Target / origin output tokens | 1 | multiplier | | vLLM / Ollama (one copy) | Target price guess · read | 0 | USD/read | | vLLM / Ollama (one copy) | Target price guess · write | 0 | USD/write | | vLLM / Ollama (one copy) | Target price guess · provider operation | 0 | USD/provider operation | | vLLM / Ollama (one copy) | Target price guess · GiB-month | 0.17 | USD/GiB-month | | vLLM / Ollama (one copy) | Target price guess · published GiB | 0 | USD/published GiB | | vLLM / Ollama (one copy) | Target price guess · delivered GiB | 0 | USD/delivered GiB | | vLLM / Ollama (one copy) | Target price guess · network GiB | 0.09 | USD/network GiB | | vLLM / Ollama (one copy) | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | vLLM / Ollama (one copy) | Target price guess · vCPU-hour | 0.05 | USD/vCPU-hour | | vLLM / Ollama (one copy) | Target price guess · GiB-hour | 0.006 | USD/GiB-hour | | vLLM / Ollama (one copy) | Target price guess · resource-hour | 0 | USD/resource-hour | | vLLM / Ollama (one copy) | Target price guess · resource-month | 0 | USD/resource-month | | vLLM / Ollama (one copy) | Target price guess · input token | 0 | USD/input token | | vLLM / Ollama (one copy) | Target price guess · output token | 0 | USD/output token | | vLLM / Ollama (one copy) | Target monthly base charge | 0 | USD/month | | vLLM / Ollama (one copy) | Approximate target service minimum | 0 | USD/month | | vLLM / Ollama (one copy) | Minimum primary CPU | 2 | vCPU | | vLLM / Ollama (one copy) | Minimum primary RAM | 8 | GiB | | vLLM / Ollama (one copy) | Minimum primary disk | 100 | GiB | | vLLM / Ollama (one copy) | Additional full-price standby copy | 0 | 0 or 1 | | vLLM / Ollama (one copy) | Allocated share of provisioned cost | 1 | fraction | | vLLM / Ollama (one copy) | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | vLLM / Ollama (one copy) | Planning peak / mean work | 2 | multiplier | | vLLM / Ollama (one copy) | Planning primary CPU utilization | 0.7 | fraction | | vLLM / Ollama (one copy) | Resident state reserved in RAM | 0.1 | fraction | | vLLM / Ollama (one copy) | Planning disk utilization | 0.8 | fraction | | vLLM / Ollama (one copy) | Maintenance and log disk reserve | 20 | GiB | | vLLM / Ollama (one copy) | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | vLLM / Ollama (one copy) | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | vLLM / Ollama (one copy) | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | vLLM / Ollama (one copy) | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | vLLM / Ollama (one copy) | Dependent object sink write price guess | 0.000005 | USD/write | | vLLM / Ollama (two copies) | Origin list price · Input tokens | 0.000003 | USD/Tokens | | vLLM / Ollama (two copies) | Origin list price · Output tokens | 0.000015 | USD/Tokens | | vLLM / Ollama (two copies) | Target / origin input tokens | 1 | multiplier | | vLLM / Ollama (two copies) | Target / origin output tokens | 1 | multiplier | | vLLM / Ollama (two copies) | Target price guess · read | 0 | USD/read | | vLLM / Ollama (two copies) | Target price guess · write | 0 | USD/write | | vLLM / Ollama (two copies) | Target price guess · provider operation | 0 | USD/provider operation | | vLLM / Ollama (two copies) | Target price guess · GiB-month | 0.17 | USD/GiB-month | | vLLM / Ollama (two copies) | Target price guess · published GiB | 0 | USD/published GiB | | vLLM / Ollama (two copies) | Target price guess · delivered GiB | 0 | USD/delivered GiB | | vLLM / Ollama (two copies) | Target price guess · network GiB | 0.09 | USD/network GiB | | vLLM / Ollama (two copies) | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | vLLM / Ollama (two copies) | Target price guess · vCPU-hour | 0.05 | USD/vCPU-hour | | vLLM / Ollama (two copies) | Target price guess · GiB-hour | 0.006 | USD/GiB-hour | | vLLM / Ollama (two copies) | Target price guess · resource-hour | 0 | USD/resource-hour | | vLLM / Ollama (two copies) | Target price guess · resource-month | 0 | USD/resource-month | | vLLM / Ollama (two copies) | Target price guess · input token | 0 | USD/input token | | vLLM / Ollama (two copies) | Target price guess · output token | 0 | USD/output token | | vLLM / Ollama (two copies) | Target monthly base charge | 0 | USD/month | | vLLM / Ollama (two copies) | Approximate target service minimum | 0 | USD/month | | vLLM / Ollama (two copies) | Minimum primary CPU | 2 | vCPU | | vLLM / Ollama (two copies) | Minimum primary RAM | 8 | GiB | | vLLM / Ollama (two copies) | Minimum primary disk | 100 | GiB | | vLLM / Ollama (two copies) | Additional full-price standby copy | 1 | 0 or 1 | | vLLM / Ollama (two copies) | Allocated share of provisioned cost | 1 | fraction | | vLLM / Ollama (two copies) | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | vLLM / Ollama (two copies) | Planning peak / mean work | 2 | multiplier | | vLLM / Ollama (two copies) | Planning primary CPU utilization | 0.7 | fraction | | vLLM / Ollama (two copies) | Resident state reserved in RAM | 0.1 | fraction | | vLLM / Ollama (two copies) | Planning disk utilization | 0.8 | fraction | | vLLM / Ollama (two copies) | Maintenance and log disk reserve | 20 | GiB | | vLLM / Ollama (two copies) | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | vLLM / Ollama (two copies) | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | vLLM / Ollama (two copies) | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | vLLM / Ollama (two copies) | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | vLLM / Ollama (two copies) | Dependent object sink write price guess | 0.000005 | USD/write | * Only this native meter varies; all other source usage and translation assumptions retain the displayed defaults. Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. Other native meters retain their explicitly displayed default quantities. * Planning estimate · editable authored rates and sizing assumptions dated 2026-10-22; not verified provider prices. us-east-1 → Explicit target planning region, USD/month. * Source prices are reference charges, not an invoice. Target lines cover selected service usage/configuration only; adapter hosting and Tensor9 fees are excluded. * Explicit tariff samples are not performance measurements or capacity qualification. No interpolation or break-even guarantee is implied. * Variant assumptions show the first axis sample; any axis-dependent configuration is described by this curve's workload note. * Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. * Default-zero rates are explicit assumptions, not evidence that the deployed service or resource is free. * Provisioned-unit minimums use an editable approximate allocation floor. Modeled target service cost is max(usage/allocation, minimum), with billable network, durable staging and dependent sinks added separately; actual RU/TU sizing, scaling increments and throughput remain unqualified. * Durable staging is priced once: provisioned targets include adapter state in their selected allocation; native targets add a separate staging allocation when required. Adapter hosting is a separate allocation and must exclude resources already priced as staging, workers or target capacity. * A declared adapter mapping and cost model do not establish implementation availability or service equivalence. * This is a declared directed mapping; a planning estimate does not establish deployment or request-serving support. * Origin unit prices are cited provider list prices; capacity coefficients and translation assumptions are editable authored planning guesses, not measured capacity or a provider quote. * Origin tariffs are the cited Claude Sonnet 4.6 global-endpoint list prices; target tariffs name no asserted current model SKU. Select model-specific rates before using the estimate. * Input and output meters and token expansion remain independent. A cheaper model is not automatically equivalent in capability or quality. * All target rates are editable authored planning guesses, not verified current list prices, customer discounts or benchmark evidence. * Dedicated accelerator/model-fit target: the Bedrock family returns this direction as Unpriced because the serving GPU or model deployment is not selected, so these generic rates are not shown as list prices. Scaleway Managed Inference lists EUR 0.93/hour for an L4-1-24G node ([https://www.scaleway.com/en/pricing/model-as-a-service/](https://www.scaleway.com/en/pricing/model-as-a-service/)); the Azure RAI policy has no meter of its own, its tokens bill on the Azure OpenAI deployment and knowledge bases on Azure AI Search (Basic USD 0.101/hour); the Tensor9 in-cluster model service needs GPU node pricing. Read 2026-10-22. * Row-store throughput (capacityOpsPerVcpuSecond, default 1500) is measured-by-tensor9 as a lower bound: the DynamoDB adapter on Cloud SQL db-custom-2-7680 (2 vCPU, PostgreSQL 16, us-west1) served 2,974 ops/s of YCSB-A at 16 threads without recorded saturation (src/main/kotlin/t9/prime/equiv/generated/eval/CloudsqlPostgresqlYcsbAT16GoYcsbV1\_0\_0.kt). Published CPU-bound pgbench reaches 3,101 read-write transactions and 20,289 point reads per vCPU-second (AWS z1d.2xlarge, PostgreSQL 14.5), so actual capacity is likely higher. Reads and writes share one coefficient: at the same adapter version, read-only YCSB-C was not faster than 50%-update YCSB-A. * Log-broker throughput (capacityStreamMiBPerVcpuSecond 20 at 0.5 utilization, 4 GiB RAM per vCPU) is provider-documented by Google Managed Service for Apache Kafka sizing ([https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size](https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size)): write-equivalent bytes are replicated writes plus a quarter of reads. Producer batches below 10 KB lower per-CPU throughput; that penalty is not modeled. * Disk utilization 0.8 is provider-documented headroom below Cloud SQL automatic storage increase, which triggers near 91-95% full ([https://docs.cloud.google.com/sql/docs/postgres/instance-settings](https://docs.cloud.google.com/sql/docs/postgres/instance-settings)). The 20 GiB reserve exceeds PostgreSQL's 1 GB default max\_wal\_size; the remainder of that reserve is unmeasured. * CPU tiers, 4 GiB per vCPU, at most 6.5 GiB per vCPU, 96 vCPU and 65,536 GiB disk follow Cloud SQL Enterprise custom machine limits (provider-documented, same page); 624 GiB is 96 × 6.5. Targets billed per node, unit or throughput unit are sized by their backend instead. * No public evidence: peak/mean work (2) and the RAM working-set fraction (0.1). Proposed measurements: the p99 one-minute rate over the monthly mean from customer CloudWatch metrics; the buffer hit ratio while stepping RAM under a fixed YCSB-B load. * Stored GiB already includes the origin family's physical footprint assumptions; no second storage amplification is applied. * The operations/vCPU coefficient supplies the compute requirement alongside explicit resource-hour minima. * Copies and allocation fractions change cost only; they do not increase modeled primary capacity. * AWS Price List API, AmazonBedrockFoundationModels offer: [https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json) Cost scope: Modeled target native service and explicitly modeled state allocation; hosting and Tensor9 fees only when supplied. Excluded costs and workload features: * Prompt caching, batch discounts, context tiers, grounding, tools, agents, embeddings, search, guardrails and dedicated inference hardware. * Taxes, discounts, credits and free-tier allowances unless explicitly reflected in edited rates. * Unmodeled dependent services, task execution and operations outside the selected workload family. List prices read: 2026-10-22. Model: `plan-952508c3bf76f8b345ea3c168617029855c8f41f` / `2026-10-22.planning.2`. Curve: `workload-1`. Sources: [AWS Price List API, AmazonBedrockFoundationModels offer](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json). ### Bedrock model inference · Output tokens per second Estimated monthly service charges by Output tokens per second. Adapter hosting and Tensor9 fees are excluded. Source: us-east-1; target: Explicit target planning region; currency: USD. ![Bedrock model inference · Output tokens per second: monthly service charges by Output tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/b8b837ee5899aa47c0712af486b1b56a2cc564325290a28d9152eb6a079cbd4e-light.svg) ![Bedrock model inference · Output tokens per second: monthly service charges by Output tokens per second. Exact values are in the following table.](/images/diagrams/service-cost/4d8b8d78c8eacf8473dab9c54ae0afd492c883809fb60ac2c3ac13ccae72d06f-dark.svg) Comparison pills describe all sampled scale points; card amounts show the named workload. Costs The Same means within 20% at every point. Costs Depend includes mixed results and unpriced samples. Exact sampled model outputs in USD/month. Difference means target minus origin; a positive difference means the target costs more. Estimated allocation from planning assumptions; not benchmark calibrated. Lines connect evaluated samples; intermediate capacity and prices are not inferred. Unpriced means no target quote or cost difference is available within the sizing policy. | Output tokens per second (tokens/s) | Bedrock (LLM inference) | Target scenario | vLLM / Ollama | Difference | | -: | -: | - | -: | -: | | 0 | 300 | vLLM / Ollama (one copy) | Unpriced | Unavailable | | 0 | 300 | vLLM / Ollama (two copies) | Unpriced | Unavailable | | 10 | 694.2 | vLLM / Ollama (one copy) | Unpriced | Unavailable | | 10 | 694.2 | vLLM / Ollama (two copies) | Unpriced | Unavailable | | 100 | 4242 | vLLM / Ollama (one copy) | Unpriced | Unavailable | | 100 | 4242 | vLLM / Ollama (two copies) | Unpriced | Unavailable | | 1000 | 39720 | vLLM / Ollama (one copy) | Unpriced | Unavailable | | 1000 | 39720 | vLLM / Ollama (two copies) | Unpriced | Unavailable | | 10000 | 394500 | vLLM / Ollama (one copy) | Unpriced | Unavailable | | 10000 | 394500 | vLLM / Ollama (two copies) | Unpriced | Unavailable | | 100000 | 3942300 | vLLM / Ollama (one copy) | Unpriced | Unavailable | | 100000 | 3942300 | vLLM / Ollama (two copies) | Unpriced | Unavailable | **Estimated allocation from planning assumptions; not benchmark calibrated.** The sizing policy grows compute with translated work and peak headroom, memory with compute and working set, and disk with resident working state and reserve. Monthly usage is distinct from resident data; each workload's assumptions define the translation. HA changes pricing, not this policy's throughput estimate. Unpriced workloads exceed the policy's configuration range; no silent vertical cap or automatic sharding is assumed. | Sample / scenario | Basis | Mean / peak work/s | Required vCPU | Allocated vCPU | Memory GiB | SSD GiB | Resident GiB | Policy | Constraints | | - | - | -: | -: | -: | -: | -: | -: | - | - | | 0 tokens/s / vLLM / Ollama (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 0 tokens/s / vLLM / Ollama (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10 tokens/s / vLLM / Ollama (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10 tokens/s / vLLM / Ollama (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100 tokens/s / vLLM / Ollama (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100 tokens/s / vLLM / Ollama (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 1000 tokens/s / vLLM / Ollama (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 1000 tokens/s / vLLM / Ollama (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10000 tokens/s / vLLM / Ollama (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 10000 tokens/s / vLLM / Ollama (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100000 tokens/s / vLLM / Ollama (one copy) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | | 100000 tokens/s / vLLM / Ollama (two copies) | Unpriced | 0 / 0 | 0 | Unpriced | Unpriced | Unpriced | 0 | `service-adapter-planning-2026-10-22.1` | Dedicated inference accelerator allocation and model fit have not been selected. | * Row-store throughput (capacityOpsPerVcpuSecond, default 1500) is measured-by-tensor9 as a lower bound: the DynamoDB adapter on Cloud SQL db-custom-2-7680 (2 vCPU, PostgreSQL 16, us-west1) served 2,974 ops/s of YCSB-A at 16 threads without recorded saturation (src/main/kotlin/t9/prime/equiv/generated/eval/CloudsqlPostgresqlYcsbAT16GoYcsbV1\_0\_0.kt). Published CPU-bound pgbench reaches 3,101 read-write transactions and 20,289 point reads per vCPU-second (AWS z1d.2xlarge, PostgreSQL 14.5), so actual capacity is likely higher. Reads and writes share one coefficient: at the same adapter version, read-only YCSB-C was not faster than 50%-update YCSB-A. * Log-broker throughput (capacityStreamMiBPerVcpuSecond 20 at 0.5 utilization, 4 GiB RAM per vCPU) is provider-documented by Google Managed Service for Apache Kafka sizing ([https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size](https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size)): write-equivalent bytes are replicated writes plus a quarter of reads. Producer batches below 10 KB lower per-CPU throughput; that penalty is not modeled. * Disk utilization 0.8 is provider-documented headroom below Cloud SQL automatic storage increase, which triggers near 91-95% full ([https://docs.cloud.google.com/sql/docs/postgres/instance-settings](https://docs.cloud.google.com/sql/docs/postgres/instance-settings)). The 20 GiB reserve exceeds PostgreSQL's 1 GB default max\_wal\_size; the remainder of that reserve is unmeasured. * CPU tiers, 4 GiB per vCPU, at most 6.5 GiB per vCPU, 96 vCPU and 65,536 GiB disk follow Cloud SQL Enterprise custom machine limits (provider-documented, same page); 624 GiB is 96 × 6.5. Targets billed per node, unit or throughput unit are sized by their backend instead. * No public evidence: peak/mean work (2) and the RAM working-set fraction (0.1). Proposed measurements: the p99 one-minute rate over the monthly mean from customer CloudWatch metrics; the buffer hit ratio while stepping RAM under a fixed YCSB-B load. * Stored GiB already includes the origin family's physical footprint assumptions; no second storage amplification is applied. * The operations/vCPU coefficient supplies the compute requirement alongside explicit resource-hour minima. * Copies and allocation fractions change cost only; they do not increase modeled primary capacity. **Curve assumptions** | Target scenario | Assumption | Value | Unit | | - | - | -: | - | | vLLM / Ollama (one copy) | Origin list price · Input tokens | 0.000003 | USD/Tokens | | vLLM / Ollama (one copy) | Origin list price · Output tokens | 0.000015 | USD/Tokens | | vLLM / Ollama (one copy) | Target / origin input tokens | 1 | multiplier | | vLLM / Ollama (one copy) | Target / origin output tokens | 1 | multiplier | | vLLM / Ollama (one copy) | Target price guess · read | 0 | USD/read | | vLLM / Ollama (one copy) | Target price guess · write | 0 | USD/write | | vLLM / Ollama (one copy) | Target price guess · provider operation | 0 | USD/provider operation | | vLLM / Ollama (one copy) | Target price guess · GiB-month | 0.17 | USD/GiB-month | | vLLM / Ollama (one copy) | Target price guess · published GiB | 0 | USD/published GiB | | vLLM / Ollama (one copy) | Target price guess · delivered GiB | 0 | USD/delivered GiB | | vLLM / Ollama (one copy) | Target price guess · network GiB | 0.09 | USD/network GiB | | vLLM / Ollama (one copy) | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | vLLM / Ollama (one copy) | Target price guess · vCPU-hour | 0.05 | USD/vCPU-hour | | vLLM / Ollama (one copy) | Target price guess · GiB-hour | 0.006 | USD/GiB-hour | | vLLM / Ollama (one copy) | Target price guess · resource-hour | 0 | USD/resource-hour | | vLLM / Ollama (one copy) | Target price guess · resource-month | 0 | USD/resource-month | | vLLM / Ollama (one copy) | Target price guess · input token | 0 | USD/input token | | vLLM / Ollama (one copy) | Target price guess · output token | 0 | USD/output token | | vLLM / Ollama (one copy) | Target monthly base charge | 0 | USD/month | | vLLM / Ollama (one copy) | Approximate target service minimum | 0 | USD/month | | vLLM / Ollama (one copy) | Minimum primary CPU | 2 | vCPU | | vLLM / Ollama (one copy) | Minimum primary RAM | 8 | GiB | | vLLM / Ollama (one copy) | Minimum primary disk | 100 | GiB | | vLLM / Ollama (one copy) | Additional full-price standby copy | 0 | 0 or 1 | | vLLM / Ollama (one copy) | Allocated share of provisioned cost | 1 | fraction | | vLLM / Ollama (one copy) | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | vLLM / Ollama (one copy) | Planning peak / mean work | 2 | multiplier | | vLLM / Ollama (one copy) | Planning primary CPU utilization | 0.7 | fraction | | vLLM / Ollama (one copy) | Resident state reserved in RAM | 0.1 | fraction | | vLLM / Ollama (one copy) | Planning disk utilization | 0.8 | fraction | | vLLM / Ollama (one copy) | Maintenance and log disk reserve | 20 | GiB | | vLLM / Ollama (one copy) | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | vLLM / Ollama (one copy) | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | vLLM / Ollama (one copy) | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | vLLM / Ollama (one copy) | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | vLLM / Ollama (one copy) | Dependent object sink write price guess | 0.000005 | USD/write | | vLLM / Ollama (two copies) | Origin list price · Input tokens | 0.000003 | USD/Tokens | | vLLM / Ollama (two copies) | Origin list price · Output tokens | 0.000015 | USD/Tokens | | vLLM / Ollama (two copies) | Target / origin input tokens | 1 | multiplier | | vLLM / Ollama (two copies) | Target / origin output tokens | 1 | multiplier | | vLLM / Ollama (two copies) | Target price guess · read | 0 | USD/read | | vLLM / Ollama (two copies) | Target price guess · write | 0 | USD/write | | vLLM / Ollama (two copies) | Target price guess · provider operation | 0 | USD/provider operation | | vLLM / Ollama (two copies) | Target price guess · GiB-month | 0.17 | USD/GiB-month | | vLLM / Ollama (two copies) | Target price guess · published GiB | 0 | USD/published GiB | | vLLM / Ollama (two copies) | Target price guess · delivered GiB | 0 | USD/delivered GiB | | vLLM / Ollama (two copies) | Target price guess · network GiB | 0.09 | USD/network GiB | | vLLM / Ollama (two copies) | Target price guess · replica GiB-month | 0 | USD/replica GiB-month | | vLLM / Ollama (two copies) | Target price guess · vCPU-hour | 0.05 | USD/vCPU-hour | | vLLM / Ollama (two copies) | Target price guess · GiB-hour | 0.006 | USD/GiB-hour | | vLLM / Ollama (two copies) | Target price guess · resource-hour | 0 | USD/resource-hour | | vLLM / Ollama (two copies) | Target price guess · resource-month | 0 | USD/resource-month | | vLLM / Ollama (two copies) | Target price guess · input token | 0 | USD/input token | | vLLM / Ollama (two copies) | Target price guess · output token | 0 | USD/output token | | vLLM / Ollama (two copies) | Target monthly base charge | 0 | USD/month | | vLLM / Ollama (two copies) | Approximate target service minimum | 0 | USD/month | | vLLM / Ollama (two copies) | Minimum primary CPU | 2 | vCPU | | vLLM / Ollama (two copies) | Minimum primary RAM | 8 | GiB | | vLLM / Ollama (two copies) | Minimum primary disk | 100 | GiB | | vLLM / Ollama (two copies) | Additional full-price standby copy | 1 | 0 or 1 | | vLLM / Ollama (two copies) | Allocated share of provisioned cost | 1 | fraction | | vLLM / Ollama (two copies) | Read/write work per vCPU-second | 1500 | operations/vCPU-second | | vLLM / Ollama (two copies) | Planning peak / mean work | 2 | multiplier | | vLLM / Ollama (two copies) | Planning primary CPU utilization | 0.7 | fraction | | vLLM / Ollama (two copies) | Resident state reserved in RAM | 0.1 | fraction | | vLLM / Ollama (two copies) | Planning disk utilization | 0.8 | fraction | | vLLM / Ollama (two copies) | Maintenance and log disk reserve | 20 | GiB | | vLLM / Ollama (two copies) | Staging database CPU unit-price guess | 0.0413 | USD/vCPU-hour | | vLLM / Ollama (two copies) | Staging database RAM unit-price guess | 0.007 | USD/GiB-hour | | vLLM / Ollama (two copies) | Staging database disk unit-price guess | 0.17 | USD/GiB-month | | vLLM / Ollama (two copies) | Dependent object sink storage price guess | 0.023 | USD/GiB-month | | vLLM / Ollama (two copies) | Dependent object sink write price guess | 0.000005 | USD/write | * Only this native meter varies; all other source usage and translation assumptions retain the displayed defaults. Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. Other native meters retain their explicitly displayed default quantities. * Planning estimate · editable authored rates and sizing assumptions dated 2026-10-22; not verified provider prices. us-east-1 → Explicit target planning region, USD/month. * Source prices are reference charges, not an invoice. Target lines cover selected service usage/configuration only; adapter hosting and Tensor9 fees are excluded. * Explicit tariff samples are not performance measurements or capacity qualification. No interpolation or break-even guarantee is implied. * Variant assumptions show the first axis sample; any axis-dependent configuration is described by this curve's workload note. * Editable authored planning guesses, not verified provider tariffs, benchmark measurements or a commercial quote. * Default-zero rates are explicit assumptions, not evidence that the deployed service or resource is free. * Provisioned-unit minimums use an editable approximate allocation floor. Modeled target service cost is max(usage/allocation, minimum), with billable network, durable staging and dependent sinks added separately; actual RU/TU sizing, scaling increments and throughput remain unqualified. * Durable staging is priced once: provisioned targets include adapter state in their selected allocation; native targets add a separate staging allocation when required. Adapter hosting is a separate allocation and must exclude resources already priced as staging, workers or target capacity. * A declared adapter mapping and cost model do not establish implementation availability or service equivalence. * This is a declared directed mapping; a planning estimate does not establish deployment or request-serving support. * Origin unit prices are cited provider list prices; capacity coefficients and translation assumptions are editable authored planning guesses, not measured capacity or a provider quote. * Origin tariffs are the cited Claude Sonnet 4.6 global-endpoint list prices; target tariffs name no asserted current model SKU. Select model-specific rates before using the estimate. * Input and output meters and token expansion remain independent. A cheaper model is not automatically equivalent in capability or quality. * All target rates are editable authored planning guesses, not verified current list prices, customer discounts or benchmark evidence. * Dedicated accelerator/model-fit target: the Bedrock family returns this direction as Unpriced because the serving GPU or model deployment is not selected, so these generic rates are not shown as list prices. Scaleway Managed Inference lists EUR 0.93/hour for an L4-1-24G node ([https://www.scaleway.com/en/pricing/model-as-a-service/](https://www.scaleway.com/en/pricing/model-as-a-service/)); the Azure RAI policy has no meter of its own, its tokens bill on the Azure OpenAI deployment and knowledge bases on Azure AI Search (Basic USD 0.101/hour); the Tensor9 in-cluster model service needs GPU node pricing. Read 2026-10-22. * Row-store throughput (capacityOpsPerVcpuSecond, default 1500) is measured-by-tensor9 as a lower bound: the DynamoDB adapter on Cloud SQL db-custom-2-7680 (2 vCPU, PostgreSQL 16, us-west1) served 2,974 ops/s of YCSB-A at 16 threads without recorded saturation (src/main/kotlin/t9/prime/equiv/generated/eval/CloudsqlPostgresqlYcsbAT16GoYcsbV1\_0\_0.kt). Published CPU-bound pgbench reaches 3,101 read-write transactions and 20,289 point reads per vCPU-second (AWS z1d.2xlarge, PostgreSQL 14.5), so actual capacity is likely higher. Reads and writes share one coefficient: at the same adapter version, read-only YCSB-C was not faster than 50%-update YCSB-A. * Log-broker throughput (capacityStreamMiBPerVcpuSecond 20 at 0.5 utilization, 4 GiB RAM per vCPU) is provider-documented by Google Managed Service for Apache Kafka sizing ([https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size](https://docs.cloud.google.com/managed-service-for-apache-kafka/docs/plan-cluster-size)): write-equivalent bytes are replicated writes plus a quarter of reads. Producer batches below 10 KB lower per-CPU throughput; that penalty is not modeled. * Disk utilization 0.8 is provider-documented headroom below Cloud SQL automatic storage increase, which triggers near 91-95% full ([https://docs.cloud.google.com/sql/docs/postgres/instance-settings](https://docs.cloud.google.com/sql/docs/postgres/instance-settings)). The 20 GiB reserve exceeds PostgreSQL's 1 GB default max\_wal\_size; the remainder of that reserve is unmeasured. * CPU tiers, 4 GiB per vCPU, at most 6.5 GiB per vCPU, 96 vCPU and 65,536 GiB disk follow Cloud SQL Enterprise custom machine limits (provider-documented, same page); 624 GiB is 96 × 6.5. Targets billed per node, unit or throughput unit are sized by their backend instead. * No public evidence: peak/mean work (2) and the RAM working-set fraction (0.1). Proposed measurements: the p99 one-minute rate over the monthly mean from customer CloudWatch metrics; the buffer hit ratio while stepping RAM under a fixed YCSB-B load. * Stored GiB already includes the origin family's physical footprint assumptions; no second storage amplification is applied. * The operations/vCPU coefficient supplies the compute requirement alongside explicit resource-hour minima. * Copies and allocation fractions change cost only; they do not increase modeled primary capacity. * AWS Price List API, AmazonBedrockFoundationModels offer: [https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json) Cost scope: Modeled target native service and explicitly modeled state allocation; hosting and Tensor9 fees only when supplied. Excluded costs and workload features: * Prompt caching, batch discounts, context tiers, grounding, tools, agents, embeddings, search, guardrails and dedicated inference hardware. * Taxes, discounts, credits and free-tier allowances unless explicitly reflected in edited rates. * Unmodeled dependent services, task execution and operations outside the selected workload family. List prices read: 2026-10-22. Model: `plan-952508c3bf76f8b345ea3c168617029855c8f41f` / `2026-10-22.planning.2`. Curve: `workload-2`. Sources: [AWS Price List API, AmazonBedrockFoundationModels offer](https://pricing.us-east-1.amazonaws.com/offers/v1.0/aws/AmazonBedrockFoundationModels/current/us-east-1/index.json). See the [service catalog](/byoc/service-adapters/catalog.md) for operation coverage, supported directions and backend limitations. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.