Cloud platforms - 50 entries
Amazon Bedrock quotas and rate limits
AWS's official docs, read on November 5, 2026, set quotas for Amazon Bedrock. For Anthropic Claude Opus 4.8, the docs list 30,000,000 Cross-region TPM [adjustable], 30,000,000 Global cross-region TPM [adjustable] and 20,000,000 Mantle input TPM [adjustable].
Last verified from docs.aws.amazon.com. What the docs leave out
- Sources
- 2
- Tables
- 35
- Entries
- 50
- Verified
- Nov 5, 2026
Limits by model, endpoint and plan
Values are copied exactly as the docs show them, units included, in the docs' own order. A dash means the docs list no value for that row.
All 35 tables on this page
- bedrock-mantle endpoint (OpenAI Responses API, OpenAI Chat Completions API, Anthropic Messages API)
- Anthropic Claude Opus 5, Anthropic Claude Opus 5.5
- Anthropic Claude Opus 4.8 and Anthropic Claude Opus 4.7
- Anthropic Claude Sonnet 5
- Anthropic Claude Sonnet 5.5 and Anthropic Claude Haiku 5.5
- Anthropic Claude Fable 5.1 and Anthropic Claude Fable 5
- Anthropic Claude Sonnet 4.6 and Anthropic Claude Opus 4.6 V1
- Anthropic Claude Opus 4.5
- Anthropic Claude Haiku 4.5, Anthropic Claude Sonnet 4.5 V1
- Anthropic Claude Sonnet 4.5 V1 1M Context Length
- Anthropic Claude 3.7 Sonnet V1
- Anthropic Claude 3.5 Haiku
- Anthropic Claude 3.5 Sonnet V2
- Anthropic Claude 3.5 Sonnet
- Anthropic Claude 3 Haiku
- Anthropic Claude 3 Sonnet
- GPT-5.6 Luna, GPT-5.6 Terra and 2 more
- GPT-6 Luna and GPT-6 Sol, GPT-6.1 Sol
- GPT-6.1 Terra
- GPT-5.4 and GPT-5.5
- Grok 4.6, Grok 4.7
- Moonshot AI Kimi K3 and Moonshot AI Kimi K3.1
- Z.ai GLM 5.3
- Amazon Nova 2 Lite, Amazon Nova 2 Omni and Amazon Nova 2 Pro Preview
- Amazon Nova Premier V1
- Amazon Nova Pro (and Amazon Nova Pro V1 latency-optimized)
- Amazon Nova Lite and Amazon Nova Micro
- DeepSeek R1 V1 and Meta Llama 4 Maverick V1, Meta Llama 4 Scout V1
- Meta Llama 3.3 70B Instruct
- Meta Llama 3.1 405B Instruct and Meta Llama 3.1 70B Instruct
- Meta Llama 3.1 8B Instruct
- Meta Llama 3.2 1B Instruct, Meta Llama 3.2 3B Instruct
- Mistral Pixtral Large 25.02 V1 and Writer AI Palmyra X4 V1, Writer AI Palmyra X5 V1
- Cohere Embed V4
- Twelve Labs Marengo, Twelve Labs Pegasus, TwelveLabs Marengo Embed 3.0, TwelveLabs Pegasus 1.5
bedrock-mantle endpoint (OpenAI Responses API, OpenAI Chat Completions API, Anthropic Messages API)
| Metric | Limit |
|---|---|
| Quota types | Bedrock Mantle input tokens per minute for ${model}; Bedrock Mantle output tokens per minute for ${model} |
| RPM | Not enforced: "The bedrock-mantle endpoint does not enforce requests-per-minute (RPM) quotas." |
| Input TPM admission check | input tokens in the request plus max_tokens (or the model-specific maximum if max_tokens is not set); if over quota the request is throttled with an HTTP 429 response |
| Output TPM | counted during generation; if reached, generation stops and the response is returned with a finish reason indicating the cutoff |
- Separate allocation from bedrock-runtime even for the same model. Cached input tokens read through prompt caching do not count against input TPM. "The endpoint may apply additional internal rate limiting that is not exposed in Service Quotas." Per-account values are in the Service Quotas console (search bedrock-mantle endpoint).
Anthropic Claude Opus 5, Anthropic Claude Opus 5.5
Model pages: Claude Opus 5, Claude Opus 5.5
| Metric | Limit |
|---|---|
| Cross-region TPM [adjustable] | 30,000,000 |
| Global cross-region TPM [adjustable] | 30,000,000 |
| On-demand TPM | 15,000,000 |
| Mantle input TPM [adjustable] | 20,000,000 |
| Mantle output TPM [adjustable] | 2,000,000 |
Anthropic Claude Opus 4.8 and Anthropic Claude Opus 4.7
2 scopes
| Model or scope | Cross-region TPM [adjustable] |
|---|---|
| Anthropic Claude Opus 4.8 | 30,000,000 |
| |
| Anthropic Claude Opus 4.7 | 10,000,000 |
| |
- The General Reference table lists no RPM quota for this model.
Anthropic Claude Sonnet 5
Model page: Claude Sonnet 5
| Metric | Limit |
|---|---|
| Cross-region TPM [adjustable] | 6,000,000 |
| Global cross-region TPM [adjustable] | 6,000,000 |
| On-demand TPM | 3,000,000 |
| Mantle input TPM [adjustable] | 3,000,000 |
| Mantle output TPM [adjustable] | 300,000 |
Anthropic Claude Sonnet 5.5 and Anthropic Claude Haiku 5.5
2 scopes
| Model or scope | Cross-region TPM [adjustable] |
|---|---|
| Anthropic Claude Sonnet 5.5 | 6,000,000 |
| |
| Anthropic Claude Haiku 5.5 | 10,000,000 |
| |
Anthropic Claude Fable 5.1 and Anthropic Claude Fable 5
2 scopes
| Model or scope | Cross-region TPM [adjustable] |
|---|---|
| Anthropic Claude Fable 5.1 | 10,000,000 |
| |
| Anthropic Claude Fable 5 | 4,000,000 |
| |
Anthropic Claude Sonnet 4.6 and Anthropic Claude Opus 4.6 V1
2 scopes
| Model or scope | Cross-region RPM [adjustable] |
|---|---|
| Anthropic Claude Sonnet 4.6 | 10,000 |
| |
| Anthropic Claude Opus 4.6 V1 | 10,000 |
| |
Anthropic Claude Opus 4.5
Model page: Claude Opus 4.5
| Metric | Limit |
|---|---|
| Cross-region RPM [adjustable] | 10,000 |
| Cross-region TPM [adjustable] | 2,000,000 |
| Global cross-region RPM [adjustable] | 10,000 |
| Global cross-region TPD | 2,880,000,000 |
| Global cross-region TPM [adjustable] | 2,000,000 |
| Max TPD (doubled for cross-region calls) | 1,440,000,000 |
Anthropic Claude Haiku 4.5, Anthropic Claude Sonnet 4.5 V1
Model page: Claude Haiku 4.5
| Metric | Limit |
|---|---|
| Cross-region RPM [adjustable] | 10,000 |
| Cross-region TPM [adjustable] | 5,000,000 |
| Global cross-region RPM [adjustable] | 10,000 |
| Global cross-region TPM [adjustable] | 5,000,000 |
Anthropic Claude Sonnet 4.5 V1 1M Context Length
| Metric | Limit |
|---|---|
| Cross-region RPM [adjustable] | 1,000 |
| Cross-region TPM [adjustable] | 1,000,000 |
| Global cross-region RPM [adjustable] | 1,000 |
| Global cross-region TPD | 1,440,000,000 |
| Global cross-region TPM [adjustable] | 1,000,000 |
| Max TPD (doubled for cross-region calls) | 720,000,000 |
Anthropic Claude 3.7 Sonnet V1
| Metric | Limit |
|---|---|
| Cross-region RPM | 50 |
| Cross-region TPM [adjustable] | 50,000 |
| Max TPD (doubled for cross-region calls) | 720,000,000 |
| On-demand RPM | 125 |
| On-demand TPM | 500,000 |
Anthropic Claude 3.5 Haiku
| Metric | Limit |
|---|---|
| Cross-region RPM | 2,000 |
| Cross-region TPM [adjustable] | 4,000,000 |
| Max TPD (doubled for cross-region calls) | 2,880,000,000 |
| On-demand latency-optimized RPM | 100 |
| On-demand latency-optimized TPM | 500,000 |
| On-demand RPM | us-west-1: 400 ; Each of the other supported Regions: 1,000 |
| On-demand TPM | us-west-1: 300,000 ; Each of the other supported Regions: 2,000,000 |
Anthropic Claude 3.5 Sonnet V2
| Metric | Limit |
|---|---|
| Cross-region RPM | 100 |
| Cross-region TPM [adjustable] | 800,000 |
| Max TPD (doubled for cross-region calls) | 576,000,000 |
| On-demand RPM | 50 |
| On-demand TPM | 400,000 |
Anthropic Claude 3.5 Sonnet
| Metric | Limit |
|---|---|
| Cross-region RPM | ap-southeast-1: 40 ; Each of the other supported Regions: 100 |
| Cross-region TPM [adjustable] | ap-southeast-1: 400,000 ; Each of the other supported Regions: 800,000 |
| On-demand RPM | ap-southeast-1: 20 ; Each of the other supported Regions: 50 |
| On-demand TPM | ap-southeast-1: 200,000 ; Each of the other supported Regions: 400,000 |
Anthropic Claude 3 Haiku
| Metric | Limit |
|---|---|
| Cross-region RPM | ap-southeast-1: 400 ; Each of the other supported Regions: 800 |
| Cross-region TPM [adjustable] | ap-southeast-1: 400,000 ; Each of the other supported Regions: 600,000 |
| Max TPD (doubled for cross-region calls) | ap-southeast-1: 288,000,000 ; Each of the other supported Regions: 432,000,000 |
| On-demand RPM | ap-southeast-1: 200 ; Each of the other supported Regions: 400 |
| On-demand TPM | ap-southeast-1: 200,000 ; Each of the other supported Regions: 300,000 |
Anthropic Claude 3 Sonnet
| Metric | Limit |
|---|---|
| Cross-region RPM | 200 |
| Cross-region TPM [adjustable] | 400,000 |
| On-demand RPM | 100 |
| On-demand TPM | 200,000 |
GPT-5.6 Luna, GPT-5.6 Terra and 2 more
4 scopes
| Model or scope | Cross-region TPM [adjustable] |
|---|---|
| GPT-5.6 Luna | 4,000,000 |
| |
| GPT-5.6 Terra | 20,000,000 |
| |
| GPT-5.6 Sol | 2,000,000 |
| |
| GPT-6 Astra | 4,000,000 |
| |
GPT-6 Luna and GPT-6 Sol, GPT-6.1 Sol
2 scopes
| Model or scope | Cross-region TPM [adjustable] |
|---|---|
| GPT-6 Luna | 4,000,000 |
| |
| GPT-6 Sol, GPT-6.1 Sol | 2,000,000 |
| |
GPT-6.1 Terra
| Metric | Limit |
|---|---|
| Cross-region TPM [adjustable] | 2,000,000 |
| Global cross-region TPD | 5,760,000,000 |
| Global cross-region TPM [adjustable] | 4,000,000 |
| Mantle input TPM [adjustable] | 1,000,000 |
| Mantle output TPM [adjustable] | 100,000 |
| Max TPD (doubled for cross-region calls) | 1,440,000,000 |
GPT-5.4 and GPT-5.5
2 scopes
- Only bedrock-mantle quotas are listed for this model.
| Metric | Limit |
|---|---|
| Cross-region TPM [adjustable] | 10,000,000 |
| Global cross-region TPM [adjustable] | 10,000,000 |
Moonshot AI Kimi K3 and Moonshot AI Kimi K3.1
2 scopes
| Model or scope | Cross-region TPM [adjustable] |
|---|---|
| Moonshot AI Kimi K3 | 10,000,000 |
| |
| Moonshot AI Kimi K3.1 | 10,000,000 |
| |
Z.ai GLM 5.3
Model page: GLM-5.3
| Metric | Limit |
|---|---|
| Cross-region TPM [adjustable] | 10,000,000 |
| Global cross-region TPM [adjustable] | 10,000,000 |
| Mantle input TPM [adjustable] (published as "GLM 5.3") | 10,000,000 |
| Mantle output TPM [adjustable] (published as "GLM 5.3") | 1,000,000 |
Amazon Nova 2 Lite, Amazon Nova 2 Omni and Amazon Nova 2 Pro Preview
2 scopes
| Model or scope | Cross-region RPM |
|---|---|
| Amazon Nova 2 Lite, Amazon Nova 2 Omni | 2,000 |
| |
| Amazon Nova 2 Pro Preview | 100 |
| |
Amazon Nova Premier V1
| Metric | Limit |
|---|---|
| Cross-region RPM | 500 |
| Cross-region TPM [adjustable] | 2,000,000 |
| Max TPD (doubled for cross-region calls) | 1,440,000,000 |
Amazon Nova Pro (and Amazon Nova Pro V1 latency-optimized)
Model page: Amazon Nova Pro
| Metric | Limit |
|---|---|
| Cross-region RPM | 500 |
| Cross-region TPM [adjustable] | 2,000,000 |
| Max TPD (doubled for cross-region calls) | 1,440,000,000 |
| On-demand RPM | 250 |
| On-demand TPM | 1,000,000 |
| Nova Pro V1: Model invocation max latency-optimized tokens per day | 57,600,000 |
| Nova Pro V1: On-Demand, latency-optimized RPM | 10 |
| Nova Pro V1: On-Demand, latency-optimized TPM | 40,000 |
- Latency-optimized rows are published under the name "Amazon Nova Pro V1".
Amazon Nova Lite and Amazon Nova Micro
2 scopes
| Model or scope | Cross-region RPM |
|---|---|
| Amazon Nova Lite | us-east-1: 4,000 ; us-east-2: 4,000 ; us-west-1: 4,000 ; us-west-2: 4,000 ; Each of the other supported Regions: 400 |
| |
| Amazon Nova Micro | us-east-1: 4,000 ; us-east-2: 4,000 ; us-west-2: 4,000 ; Each of the other supported Regions: 400 |
| |
DeepSeek R1 V1 and Meta Llama 4 Maverick V1, Meta Llama 4 Scout V1
2 scopes
| Model or scope | Cross-region RPM |
|---|---|
| DeepSeek R1 V1 | 200 |
| |
| Meta Llama 4 Maverick V1, Meta Llama 4 Scout V1 | 800 |
| |
Meta Llama 3.3 70B Instruct
| Metric | Limit |
|---|---|
| Cross-region RPM | 800 |
| Cross-region TPM [adjustable] | 600,000 |
| On-demand RPM | 400 |
| On-demand TPM | 300,000 |
Meta Llama 3.1 405B Instruct and Meta Llama 3.1 70B Instruct
2 scopes
| Model or scope | Cross-region RPM |
|---|---|
| Meta Llama 3.1 405B Instruct | 400 |
| |
| Meta Llama 3.1 70B Instruct | 800 |
| |
Meta Llama 3.1 8B Instruct
| Metric | Limit |
|---|---|
| Cross-region RPM | 1,600 |
| Cross-region TPM [adjustable] | 600,000 |
| On-demand RPM | 800 |
| On-demand TPM | 300,000 |
Meta Llama 3.2 1B Instruct, Meta Llama 3.2 3B Instruct
| Metric | Limit |
|---|---|
| Cross-region RPM | 1,600 |
| Cross-region TPM [adjustable] | 600,000 |
| Max TPD (doubled for cross-region calls) | 432,000,000 |
| On-demand RPM | 800 |
| On-demand TPM | 300,000 |
Mistral Pixtral Large 25.02 V1 and Writer AI Palmyra X4 V1, Writer AI Palmyra X5 V1
2 scopes
| Model or scope | Cross-region RPM |
|---|---|
| Mistral Pixtral Large 25.02 V1 | 10 |
| |
| Writer AI Palmyra X4 V1, Writer AI Palmyra X5 V1 | 10 |
| |
Cohere Embed V4
| Metric | Limit |
|---|---|
| Cross-region RPM | 2,000 |
| Cross-region TPM [adjustable] | 300,000 |
| Global cross-region RPM [adjustable] | 2,000 |
| Global cross-region TPD | 432,000,000 |
| Global cross-region TPM [adjustable] | 300,000 |
| Max TPD (doubled for cross-region calls) | 216,000,000 |
| On-demand RPM | 1,000 |
| On-demand TPM | 150,000 |
Twelve Labs Marengo, Twelve Labs Pegasus, TwelveLabs Marengo Embed 3.0, TwelveLabs Pegasus 1.5
| Metric | Limit |
|---|---|
| Twelve Labs Marengo: Cross-region RPM | 200 |
| Twelve Labs Marengo: On-demand InvokeModel concurrent requests | 30 |
| Twelve Labs Marengo: On-demand RPM | 100 |
| Twelve Labs Pegasus: Cross-region RPM [adjustable] | 120 |
| Twelve Labs Pegasus: Global cross-region RPM [adjustable] | 240 |
| Twelve Labs Pegasus: On-demand InvokeModel concurrent requests [adjustable] | 30 |
| Twelve Labs Pegasus: On-demand RPM [adjustable] | 60 |
| TwelveLabs Marengo Embed 3.0: Cross-region RPM [adjustable] | 1,000 |
| TwelveLabs Marengo Embed 3.0: On-demand InvokeModel async concurrent requests [adjustable] | 10 |
| TwelveLabs Marengo Embed 3.0: On-demand RPM [adjustable] | 500 |
| TwelveLabs Pegasus 1.5: Cross-region RPM [adjustable] | 60 |
| TwelveLabs Pegasus 1.5: Global cross-region RPM [adjustable] | 60 |
| TwelveLabs Pegasus 1.5: On-demand InvokeModel concurrent requests [adjustable] | 30 |
How to raise your limits
bedrock-mantle endpoint (OpenAI Responses API, OpenAI Chat Completions API, Anthropic Messages API)
bedrock-mantle quotas are visible in Service Quotas but "quota increase requests are not currently processed through the Service Quotas console". Submit the AWS Support limit increase form, select Amazon Bedrock, and specify the endpoint (bedrock-mantle), Region, model, quota name (input TPM or output TPM) and requested value. Approval depends on whether existing usage justifies the increase; include recent CloudWatch or Service Quotas usage.
Anthropic Claude Opus 5, Anthropic Claude Opus 5.5, Anthropic Claude Opus 4.8 and 47 more
runtime: Service Quotas; mantle: AWS Support limit increase form
What the docs leave out
- Per-account applied quota values: the General Reference table lists default values; the bedrock-mantle quotas page directs users to the Service Quotas console "To view the current default and applied quota values for models available to your account".
- Bedrock-mantle per-model defaults are published in the General Reference only for the models listed there; the docs say mantle quota values "vary by model, account, and AWS Region" and direct users to the Service Quotas console.
- On-demand concurrency quotas are listed only for some models: the Twelve Labs models (recorded in their entry) and Amazon Nova Sonic, Amazon Nova 2 Sonic, Amazon Nova Reel, Amazon Nova 2 Multimodal Embeddings V1 and Luma Ray V2 (not listed here). No other model listed here has one. Concurrency quotas for jobs such as evaluation, data automation and knowledge base ingestion are not listed here.
- Batch inference, custom model, imported model, Provisioned Throughput, Agents, Knowledge Bases, Guardrails and other non-inference control-plane quotas were left out of this per-model inference extraction.
- 2026-11-05: the GPT-5.6 Luna and Terra entry was split into separate GPT-5.6 Luna and GPT-5.6 Terra entries because the General Reference now lists different defaults for the two models.
FAQ
What are the Amazon Bedrock quotas and rate limits?
AWS's official docs, read on November 5, 2026, set quotas for Amazon Bedrock. For Anthropic Claude Opus 4.8, the docs list 30,000,000 Cross-region TPM [adjustable], 30,000,000 Global cross-region TPM [adjustable] and 20,000,000 Mantle input TPM [adjustable].
How do I raise my Amazon Bedrock quotas?
bedrock-mantle quotas are visible in Service Quotas but "quota increase requests are not currently processed through the Service Quotas console". Submit the AWS Support limit increase form, select Amazon Bedrock, and specify the endpoint (bedrock-mantle), Region, model, quota name (input TPM or output TPM) and requested value. Approval depends on whether existing usage justifies the increase; include recent CloudWatch or Service Quotas usage. runtime: Service Quotas; mantle: AWS Support limit increase form
Where do these Amazon Bedrock numbers come from?
Every value is copied verbatim from AWS's official documentation at docs.aws.amazon.com, read on November 5, 2026. This page lists all 50 entries in 35 tables, each linked to the page it came from.
Related reading
- AI API 429, 529 and 503 Errors: Retry Rules That Match the DocsWhich 429, 529 and 503 errors to retry across OpenAI, Claude, Gemini and Bedrock, how to read Retry-After, and a typed wrapper that skips quota errors.
- How to Use Claude Fable 5: Every Access Path ExplainedHow to use Claude Fable 5 via claude.ai plans, the Claude API, Amazon Bedrock, Vertex AI and Microsoft Foundry, with model IDs, setup effort and prompt tips.
- Self-Hosting Claude Code: 5 Ways to Run It on Your Own InfraRun Claude Code through your own gateway, Bedrock, Google Cloud or local GPUs. Five patterns with verified env vars, model IDs and what Anthropic supports.
- Which AI APIs Train on Your Data? Free Tiers, Paid APIs and Retention, Checked Against 18 Providers' Own Terms (October 2026)Free tiers are where training happens: Google's unpaid quota outside the EEA, Switzerland and the UK, Mistral's Free mode, and some OpenRouter free models. Paid APIs mostly say no. Retention is a separate question, and four providers' own docs disagree with themselves.
- Claude Code Fast Mode: Is 2.5x Speed Worth 2x Price?Claude Code fast mode pricing: $8/$40 per MTok on Opus 5.5, the first-enable context charge, separate rate limit pools, and when 2.5x speed pays off.
- Claude Fable 5 API: Production Integration Patterns, Rate Limits, and Migration GotchasClaude Fable 5 API production guide - adaptive thinking, rate limit strategy, streaming latency, fallback chains and a migration checklist from Opus 4.8.
Sources
Last verified . Every value is copied verbatim from the official pages below, read on that date.
Newsletter
Know when limits change
One email per week. New model releases, pricing and limit changes, and what they mean for what you build.