Skip to content
throttle
Menu
HomeAPI LimitsAnthropic API

AI and Machine Learning

Anthropic API

Claude model APIs for messages, tool use, batches, token counting, and related AI workloads.

PUBLISHED PROVIDER LIMITS

Limits keep their provider-specific dimensions.

Limits vary by usage tier and model class. Short bursts can be constrained more tightly than a simple per-minute average.

Authentication: Anthropic API key

Primary scope: Organization With Optional Workspace Limits

Measured dimensions: Requests Per Minute, Input Tokens Per Minute, Output Tokens Per Minute

Profile coverage: Standard Messages API request and token dimensions, tier spend caps, rate-limit headers, and spend-versus-rate error handling. Message Batches, Managed Agents, Priority Tier, fast mode, and account-specific numeric rate tables are outside this profile.

Rate limits

Anthropic API Rate limits
LimitValuePeriodScopeConditions and variability
Messages API rate limitsProvider-specificProvider-definedOrganization And Model ClassVaries by usage tier, model class, workspace configuration, and account-specific limits. Read configured values from the Claude Console or Rate Limits API.

Spend limits

Anthropic API Spend limits
LimitValuePeriodScopeConditions and variability
Start tier monthly spend cap500 USDPer calendar monthOrganizationPublished provider limit.
Build tier monthly spend cap1,000 USDPer calendar monthOrganizationPublished provider limit.
Scale tier monthly spend cap200,000 USDPer calendar monthOrganizationPublished provider limit.

RESPONSE METADATA

Headers expose the applicable limit state.

Retry-AfterFormat: Seconds
Seconds to wait before retrying a temporary rate-limit response, when present. Tier spend-cap 429 responses omit this header.
Anthropic-RateLimit-Requests-LimitFormat: Integer
Maximum requests allowed within the active request rate-limit period.
Anthropic-RateLimit-Requests-RemainingFormat: Integer
Requests remaining before being rate limited.
Anthropic-RateLimit-Requests-ResetFormat: Rfc3339
Time when request capacity will be fully replenished.
Anthropic-RateLimit-Tokens-LimitFormat: Integer
Maximum combined token allowance within the active period when applicable.
Anthropic-RateLimit-Tokens-RemainingFormat: Integer
Tokens remaining, rounded to the nearest thousand.
Anthropic-RateLimit-Tokens-ResetFormat: Rfc3339
Time when token capacity will be fully replenished.
Anthropic-RateLimit-Input-Tokens-LimitFormat: Integer
Maximum input tokens allowed within the active period.
Anthropic-RateLimit-Input-Tokens-RemainingFormat: Integer
Input tokens remaining, rounded to the nearest thousand.
Anthropic-RateLimit-Input-Tokens-ResetFormat: Rfc3339
Time when input-token capacity will be fully replenished.
Anthropic-RateLimit-Output-Tokens-LimitFormat: Integer
Maximum output tokens allowed within the active period.
Anthropic-RateLimit-Output-Tokens-RemainingFormat: Integer
Output tokens remaining, rounded to the nearest thousand.
Anthropic-RateLimit-Output-Tokens-ResetFormat: Rfc3339
Time when output-token capacity will be fully replenished.

WHEN A LIMIT IS EXCEEDED

Read the response before choosing a delay.

  • Rate-limit responses can use HTTP 429.
  • Rate and acceleration limits can return 429 with Retry-After. A tier spend-cap 429 has no Retry-After and cannot be resolved by retrying; access resumes after the cap resets or increases. User-configured organization or workspace spend limits can return HTTP 400 with invalid_request_error.
  • Error.details.error Code: enforced_spend_limit_reached identifies a Messages API tier spend-cap 429.

RECOMMENDED HANDLING

Protect both the provider and your own queue.

  • Honor Retry-After when present on temporary rate-limit responses.
  • Stop retrying enforced_spend_limit_reached; review the spend cap or wait until the next UTC calendar month.
  • For a user-configured spend-limit error, adjust the limit or wait for the stated reset.
  • Ramp traffic gradually and use prompt caching; cache reads do not count toward standard ITPM.
  • Read account-specific limits from the Console or Rate Limits API.

VERIFICATION RECORD

Official sources reviewed