Claude API Rate Limits

Q: How long do I wait after a Claude API 429 error?

Wait the number of seconds in the retry-after header; if a Claude API 429 has no retry-after and says enforced_spend_limit_reached, access returns only at 00:00 UTC on the first of next month or after a tier increase.

Claude API rate limits are set per organization by usage tier (Start, Build, Scale, Custom) as requests, input tokens and output tokens per minute; on the Start tier Claude Opus 5.5 allows 4,000 requests, 2,000,000 input tokens and 400,000 output tokens a minute. Going over returns a 429 error, while a 529 error means the API itself is overloaded.

The Claude API limits each organization in two ways: a monthly spend cap set by its usage tier ($500 on Start, $1,000 on Build, $200,000 on Scale, none on Custom) and rate limits per model in requests per minute (RPM), input tokens per minute (ITPM) and output tokens per minute (OTPM). On the Start tier, Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5 each allow 4,000 RPM, 2,000,000 ITPM and 400,000 OTPM (as of Nov 6, 2026). Exceeding a limit returns HTTP 429; a 529 is different and means the API is overloaded for everyone.

These limits apply to the developer Claude API only. The five-hour and weekly limits of the Claude apps and Claude Code subscriptions are explained in Claude Usage Limits.

Quick facts

Tiers	Evaluation (new accounts, lower limits), Start, Build, Scale, Custom (Rate limits)
Monthly spend caps	Start $500, Build $1,000, Scale $200,000; Custom none
See your limits	Console Rate limits page (`platform.claude.com/usage/limits`)
Raise them	"Request tier increase" on that page, or Anthropic support
Algorithm	Token bucket: capacity refills continuously, not at fixed minutes
Cached input	Cache reads do not count toward ITPM

What are the Claude API usage tiers?

Anthropic places each organization on a tier automatically from its usage history and account standing, and moves it up over time (Rate limits). New organizations may begin in an Evaluation tier with lower limits, which Anthropic describes as part of its fraud and abuse prevention; those limits rise on their own as history builds. Anthropic's current page lists no deposit or credit purchase that unlocks each tier, so the Console's Rate limits page is the place to see where you stand.

Message limits by tier, per model class (requests / input tokens / output tokens per minute), as of Nov 6, 2026:

Model	Start	Build	Scale
Claude Fable 5.x	4,000 / 500,000 / 100,000	10,000 / 1,500,000 / 300,000	13,000 / 4,000,000 / 800,000
Opus 5.5, Sonnet 5.5, Haiku 5.5 (each)	4,000 / 2,000,000 / 400,000	10,000 / 5,000,000 / 1,000,000	13,000 / 10,000,000 / 2,000,000
Opus 5, Opus 4.x, Sonnet 5, Sonnet 4.x, Haiku 4.5	1,000 / 2,000,000 / 400,000	5,000 / 5,000,000 / 1,000,000	10,000 / 10,000,000 / 2,000,000

Source: Rate limits, Anthropic docs, read Nov 6, 2026.

Some rows share one bucket: Fable 5.1 and Fable 5 share a limit, as do Opus 4.8, 4.7, 4.6 and 4.5, and Sonnet 4.6 and 4.5. Opus 5.5, Opus 5, Sonnet 5.5 and Sonnet 5 each have their own. Above Scale, the Custom tier is arranged with Anthropic's sales team.

Other limits: the Message Batches API allows 1,000 to 4,000 RPM by tier and 200,000 to 500,000 queued batch requests; Managed Agents allows 300 create and 1,200 read requests a minute; fast mode on Opus 5.5 has its own separate limits with anthropic-fast-* headers (Rate limits).

How do the token limits count?

Three details make the limits larger than they look (Rate limits):

Cached tokens are free for ITPM. Only uncached input and cache writes count; cache reads do not. Anthropic's example: with 2,000,000 ITPM and an 80% cache hit rate, an organization can process 10,000,000 input tokens a minute. This makes Prompt Caching the main way to raise throughput without asking for more.
max_tokens does not count. OTPM counts tokens as they are actually generated, so a high max_tokens costs nothing in rate limit terms.
Short windows still apply. A per-minute limit may be enforced per second (60 RPM as 1 a second), so bursts can fail even under the minute's total.

US-only and global inference (inference_geo) draw from the same pool.

What does a 429 error mean?

A 429 rate_limit_error has three possible causes (Claude API errors):

A rate limit. The message names which limit, and a retry-after header says how many seconds to wait.
Your tier's monthly spend cap. The same error type, but with error.details.error_code set to enforced_spend_limit_reached and no retry-after. Retries fail until 00:00 UTC on the first of the next month, unless the tier is raised (Rate limits).
Acceleration limits. A sharp jump in traffic can trigger 429s even under your limits; Anthropic advises ramping up gradually.

A spend limit you set yourself (Console > Billing > Spend limits) returns a 400 invalid_request_error beginning "You have reached your specified API usage limits", not a 429.

Every response carries headers to watch before hitting a wall: anthropic-ratelimit-requests-remaining, anthropic-ratelimit-input-tokens-remaining, anthropic-ratelimit-output-tokens-remaining and matching -limit and -reset headers (reset times in RFC 3339).

What does a 529 error mean?

A 529 overloaded_error means the API is temporarily overloaded across all users, not that your organization did anything wrong (Claude API errors). It can also arrive inside a stream that already started with HTTP 200. Retry with backoff, and check Claude Status and Outages if it persists.

How to handle limits in code

The official SDKs retry connection errors, 429s and 5xx errors with exponential backoff twice by default and honor retry-after (Claude API errors). Raise the count when you need more patience:

import anthropic

client = anthropic.Anthropic(max_retries=5)

Beyond retries, the levers are caching repeated context, moving non-urgent work to the Message Batches API (50% cheaper, with separate limits), spreading load across models that have their own buckets, and setting lower limits per workspace so one project cannot starve another.

How do I raise my Claude API rate limits?

Open the Console's Rate limits page.
Click Request tier increase and describe what you need.
For urgent needs, contact Anthropic support.

On Claude Platform on AWS that button is missing: tiers rise as paid AWS Marketplace invoices build up, or through your Anthropic account team (Rate limits). Amazon Bedrock and Google managed Vertex AI set their own quotas; see Claude on Cloud Platforms.

Workspace limits

Admins can give each workspace lower spend and rate limits than the organization's, never higher: spend limits on the Workspaces page, rate limits on the Rate limits page by picking the workspace and clicking Edit next to a model (Workspaces). The Default Workspace cannot have a spend limit, and organization-wide limits always win. The Claude Code workspace is rate-limited separately and is the only one with per-user monthly spend limits. See Claude Console.

Frequently asked questions

What are the Claude API rate limits on the free tier? The Claude API has no free tier with standing limits; new organizations get a small amount of free credit and may start in the Evaluation tier, whose limits are below the Start tier's and rise automatically with usage history (Pricing; Rate limits).

What is the difference between Claude API error 429 and 529? A Claude API 429 means your organization hit a rate limit, its monthly spend cap or an acceleration limit; a 529 means the Claude API is overloaded for all users and is not caused by your usage.

Do cached tokens count toward Claude API rate limits? No. Claude API input token limits count only uncached input and cache writes, so prompt cache reads do not use up ITPM.

How long do I wait after a Claude API 429 error? Wait the number of seconds in the retry-after header; if a Claude API 429 has no retry-after and says enforced_spend_limit_reached, access returns only at 00:00 UTC on the first of next month or after a tier increase.

Do Claude Pro or Max plans raise Claude API rate limits? No. Claude Pro and Max limits apply to the Claude apps and Claude Code; the Claude API is billed and rate-limited separately, though Max, Ultra and Team plans include monthly API credits (API credits).