Bedrock (LLM inference) costs

Tensor9 BYOC planning comparisons for aws::1.0.0::bedrock, with backend choices shown together. Read how Tensor9 estimates service-adapter costs for price scope and capacity qualification, or return to additional services.

​

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. 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. Curve assumptions

• 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
• Claude API 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, Claude API 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. 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. Curve assumptions

• 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
• Claude API 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, Claude API 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. 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. 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.

• 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): 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). 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

• 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/); 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): 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). 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

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.

​

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. 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. 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.

• 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): 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). 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

• 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/); 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): 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). 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

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.

​

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. 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. Curve assumptions

• 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). 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
• OCI price list API: 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, OCI price list API.

​

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. 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. Curve assumptions

• 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). 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
• OCI price list API: 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, OCI price list API.

​

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. 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. Curve assumptions

• 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
• OpenAI API 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, OpenAI API 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. 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. Curve assumptions

• 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
• OpenAI API 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, OpenAI API 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. 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. 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.

• 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): 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). 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

• 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/); 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): 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). 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

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.

​

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. 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. 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.

• 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): 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). 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

• 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/); 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): 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). 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

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.

​

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. 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. Curve assumptions

• 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
• 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, 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. 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. Curve assumptions

• 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
• 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, 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. 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. Curve assumptions

• 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
• 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, 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. 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. Curve assumptions

• 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
• 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, 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. 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. Curve assumptions

• 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
• 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, 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. 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. Curve assumptions

• 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
• 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, 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. 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. 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.

• 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): 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). 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

• 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/); 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): 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). 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

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.

​

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. 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. 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.

• 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): 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). 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

• 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/); 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): 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). 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

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. See the service catalog for operation coverage, supported directions and backend limitations.