Amazon Bedrock quotas and limits: how tokens count, why you get throttled, how to get more.
Bedrock limits inference per model and per Region, mostly by tokens per minute, plus a daily token cap per account and, for some models, requests per minute. Each request reserves its input tokens plus max_tokens up front, and some models count output tokens 5 to 15 times. Cut max_tokens, use cross-Region inference, then ask for an increase.
- On the main bedrock-runtime endpoint, each model has a tokens-per-minute (TPM) quota per Region for on-demand calls and another for cross-Region inference, plus a cross-model tokens-per-day quota per account. Some models also have a requests-per-minute (RPM) quota. The bedrock-mantle endpoint has its own, separate quotas.
- At the start of each request Bedrock deducts your input tokens plus max_tokens. At the end it settles to input tokens, plus cache-write tokens, plus output tokens times the model's burndown rate (1x for most models, 5x for most Claude models, 10x or 15x for some newer ones), and refunds the rest. Cache-read tokens do not count. A high max_tokens is the most common cause of surprise throttling.
- Fix order: right-size max_tokens, use prompt caching, add cross-Region inference, then request an increase in Service Quotas. AWS says increases are not automatic and go first to accounts already using their quota. If a partner builds the production setup, the partner can file POC or IW funding when it qualifies and AWS approves.
IWhat Bedrock quotas are
Bedrock limits how much you can send to each model, in each Region, per minute and per day.
Your account has default quotas that you can see in the Service Quotas console under Amazon Bedrock. AWS may set your values lower than the published defaults based on Region, account age, usage history and payment history, so always check your own account rather than a blog's table.
Bedrock has two inference endpoints, bedrock-runtime and bedrock-mantle, each with its own quotas, even for the same model. On bedrock-runtime, one model's quota is shared across every API you call it with (InvokeModel, Converse, Responses, Chat Completions).
| Quota | Scope | What it limits |
|---|---|---|
| On-demand InvokeModel tokens per minute | Per model, per Region | Input plus output tokens per minute in a single Region |
| Cross-Region InvokeModel tokens per minute | Per model, per Region | Input plus output tokens per minute through a cross-Region inference profile |
| Cross-Model Max Tokens Per Day | Per account, per Region | Tokens per day across all supported models |
| InvokeModel requests per minute | Per model, per Region | Requests per minute, only for some models |
IIHow tokens are counted against your quota
Quota use is not the same as your bill. Two rules explain most throttling.
- Rule 1: max_tokens is reserved up front: When a request starts, Bedrock deducts input tokens plus max_tokens from your TPM and daily quotas. A request with max_tokens of 32,000 reserves 32,000 output tokens even if it writes 1,000. Unused tokens come back at the end, but meanwhile fewer requests fit in the minute.
- Rule 2: output tokens burn down at a model rate: At the end, quota use is input tokens plus cache-write tokens plus output tokens times the burndown rate. On 10 October 2026 the AWS page listed 15x for Claude 4.8 models, 10x for Claude Opus 5.5, Sonnet 5, Opus 5 and Fable 5.1, 5x for other Claude models up to 4.7, 10x for OpenAI GPT-5.6 models on bedrock-runtime, and 1x for all other models.
- Cache reads are free against quota: Tokens read from the prompt cache do not count toward TPM or the daily quota. Caching a long, stable prefix raises your effective throughput.
AWS's own example: with Claude Sonnet 4 (5x burndown), a request with 1,000 input tokens and a 100-token answer uses 1,500 tokens of quota (1,000 plus 100 times 5) but is billed for 1,100. See how tokens are counted.
IIIWhat throttling looks like, and why it happens below your average
Quotas are per minute. Averages hide bursts.
When a request would push you over a quota, Bedrock rejects it with a ThrottlingException (HTTP 429). You can be well under your average and still be throttled, because a burst in one minute, a few requests with a large max_tokens, or a model with a high output burndown can fill the minute's quota.
Watch the Bedrock automatic dashboard in Amazon CloudWatch: InputTokenCount, OutputTokenCount, CacheReadInputTokenCount and CacheWriteInputTokenCount by model. Compare peak-minute usage, after burndown, to your quota.
IVHow to request a quota increase
Increases are not automatic, and AWS prioritizes accounts that already use their quota.
- Check the model's lifecycle first: AWS does not grant increases for models in Legacy or Deprecated status. Move to the successor model instead.
- Find the quota in Service Quotas: Filter to Amazon Bedrock and find the quota for your model. If the Adjustable column says yes, request a new value there.
- Ask for the TPM and daily quotas together: Request an increase on the Cross-Region InvokeModel tokens per minute quota for the model. AWS says support will then offer to raise the on-demand TPM and the cross-model daily quota too.
- Show real usage: AWS says that, due to demand, priority goes to customers whose traffic already consumes their existing quota, and requests can be denied otherwise. Bring measured peak numbers.
- Ask well before a launch: Do not plan a launch around a quota you do not have yet.
VThe levers when you hit the limit
Try the cheap ones first.
| Lever | What it does | Cost or catch |
|---|---|---|
| Lower max_tokens | Shrinks the up-front reservation so more requests fit per minute | Free; set it near your real answer length |
| Prompt caching | Cache reads do not count against quota | Cache writes can cost more; see the caching page |
| Cross-Region inference | Uses a separate cross-Region TPM quota and spreads calls across Regions in a geography | Data may be processed in another Region in that geography |
| A smaller model for easy calls | Moves volume off a high-burndown model | Check quality with an evaluation |
| Batch inference | Moves bulk work off the real-time path, with its own quotas | Not real-time; caching does not apply |
| Quota increase | Raises the ceiling | Not automatic; needs real usage |
| Provisioned Throughput | Reserved capacity for one model | Hourly cost whether used or not |
VIDesigning your client for limits
Assume you will be throttled sometimes, and make it harmless.
- Retry with exponential backoff and jitter: The AWS SDKs retry throttled calls; tune the retry mode and cap total wait time for user-facing paths.
- Put a queue in front of bursty producers: Smooth spikes from jobs, webhooks or fan-out so they do not all hit the same minute.
- Cap concurrency on the client: A limit on in-flight requests per model keeps you under TPM more reliably than retries alone.
- Count tokens before you send: Bedrock has a token-counting API, so you can estimate quota use for large prompts.
VIIWho can file AWS funding for a Bedrock build
There are two kinds of AWS money for a Bedrock build. A partner files one. You apply for the other yourself.
Partner-filed funding. An AWS consulting partner can file POC (proof of concept) funding to test a Bedrock workload, whether or not you run on AWS today. If you already run on AWS and are adding new AI work, the partner can file incremental workload (IW) funding instead. For startups this is often $50K to $150K. Depends on the workload; AWS approves. The work has to be real, and the partner has to be the one doing it.
AWS Activate, which you apply for yourself. Activate Founders gives $1,000 to start, up to $5,000, with no investor needed. Activate Portfolio gives up to $200,000 if your VC, accelerator or angel group is an Activate Provider and gives you its Org ID. A consulting partner cannot apply for Activate on your behalf. AWS made Activate credits usable for third-party models on Bedrock in April 2024.
For most startups without a VC, the partner route is the larger of the two. A partner who has run Bedrock in production can also help you design for quotas before launch. CloudRoute matches you with one vetted AWS partner for free. Partners pay CloudRoute a monthly membership plus 10% of AWS-approved funding.
| Funding | Who files it | Amount | When it fits |
|---|---|---|---|
| Proof of concept (POC) funding | Your matched AWS partner | $50K to $150K for startups. Depends on the workload; AWS approves. | You want to test something, whether or not you are on AWS today. |
| Incremental workload (IW) funding | Your matched AWS partner | $50K to $150K for startups. Depends on the workload; AWS approves. | You are already on AWS and are launching new work: a new AI feature, a new customer or a new contract. |
| AWS Activate Founders | Founder applies | $1,000 to start, up to $5,000 | Self-funded startups with no investor or accelerator behind them. |
| AWS Activate Portfolio | Founder applies | up to $200,000 | Startups backed by a VC firm, accelerator or angel investor that is an AWS Activate Provider. |
| AWS Credits for AI Startups | Invite only (AWS account manager) | $200,000 or more, by invitation | AI startups ready to scale after Activate Portfolio, invited by AWS. |
How a match works (example)
- You tell us about the work
You fill in a short form: what you run today, what you want to move, test or build, your stage and your region.
- We pick one partner
We read your answers and pick one vetted AWS consulting partner that fits the work, the stack and the region. We reply within 24 hours; your matched partner reaches out within 5 business days.
- The partner checks POC funding
Your matched partner scopes the test with you. Then it checks whether the test qualifies for POC (proof of concept) funding. A POC can fit whether or not you run on AWS today.
- The partner files, AWS decides
If the work qualifies, the partner files the funding request with AWS. AWS reviews it and decides. The amount depends on the workload.
- You pay nothing
CloudRoute is free for startups. Partners pay CloudRoute a monthly membership plus 10% of AWS-approved funding.
AWS Activate credits are separate, and you apply for them yourself. Use Activate Founders if no investor backs you. Use Activate Portfolio with the Org ID or invite from your VC, accelerator, incubator or angel group if it is an Activate Provider. A consulting partner cannot apply for Activate on your behalf.
Common questions
What are Amazon Bedrock service quotas?
Why am I throttled when my usage is below the quota?
What is the output token burndown rate?
Do cached tokens count against Bedrock quotas?
How do I request a Bedrock quota increase?
How do cross-Region inference and Provisioned Throughput help?
Can AWS funding pay for a Bedrock build?
Get a partner who can build it and file the funding
Tell us what you are building on Bedrock. We match you with one vetted AWS partner. If the work qualifies, the partner files POC funding for a test, or incremental workload (IW) funding for new work on AWS: often $50K to $150K for startups. Depends on the workload; AWS approves. CloudRoute is free for startups.