specification: API Commons Rate Limits
specificationVersion: '0.1'
schema: https://raw.githubusercontent.com/api-evangelist/interface-research/main/schema/api-commons.yml#/$defs/RateLimits
provider: Anthropic
providerId: anthropic
created: '2026-05-04'
# Provenance stamped 2026-08-11: this artifact was written by the API Evangelist
# bulk sweep dated 2026-05-04, not harvested from the provider. See roadmap#35.
method: generated
modified: '2026-05-22'
reconciled: true
tags:
  - AI
  - Rate Limiting
  - Quotas
description: Reconciled rate limits for the Anthropic Messages, Batches, and Managed Agents APIs. Token-bucket
  algorithm; only uncached input tokens count toward ITPM on most models.
sources:
  - https://www.anthropic.com/pricing
  - https://platform.claude.com/docs/en/docs/about-claude/pricing
  - https://platform.claude.com/docs/en/api/rate-limits
headers:
  limit: anthropic-ratelimit-requests-limit
  remaining: anthropic-ratelimit-requests-remaining
  reset: anthropic-ratelimit-requests-reset
  retryAfter: retry-after
  tokensLimit: anthropic-ratelimit-tokens-limit
  inputTokensLimit: anthropic-ratelimit-input-tokens-limit
  outputTokensLimit: anthropic-ratelimit-output-tokens-limit
responseCodes:
  throttled: 429
  quotaExceeded: 429
algorithm: token-bucket
spendLimits:
  - tier: Tier 1
    creditPurchase: $5
    monthlyCap: $500
  - tier: Tier 2
    creditPurchase: $40
    monthlyCap: $500
  - tier: Tier 3
    creditPurchase: $200
    monthlyCap: $1,000
  - tier: Tier 4
    creditPurchase: $400
    monthlyCap: $200,000
  - tier: Monthly Invoicing
    creditPurchase: N/A
    monthlyCap: no limit
limits:
  - tier: Tier 1
    model: Claude Opus 4.x
    rpm: 50
    itpm: 500000
    otpm: 80000
  - tier: Tier 1
    model: Claude Sonnet 4.x
    rpm: 50
    itpm: 30000
    otpm: 8000
  - tier: Tier 1
    model: Claude Haiku 4.5
    rpm: 50
    itpm: 50000
    otpm: 10000
  - tier: Tier 2
    model: Claude Opus 4.x
    rpm: 1000
    itpm: 2000000
    otpm: 200000
  - tier: Tier 2
    model: Claude Sonnet 4.x
    rpm: 1000
    itpm: 450000
    otpm: 90000
  - tier: Tier 2
    model: Claude Haiku 4.5
    rpm: 1000
    itpm: 450000
    otpm: 90000
  - tier: Tier 3
    model: Claude Opus 4.x
    rpm: 2000
    itpm: 5000000
    otpm: 400000
  - tier: Tier 3
    model: Claude Sonnet 4.x
    rpm: 2000
    itpm: 800000
    otpm: 160000
  - tier: Tier 3
    model: Claude Haiku 4.5
    rpm: 2000
    itpm: 1000000
    otpm: 200000
  - tier: Tier 4
    model: Claude Opus 4.x
    rpm: 4000
    itpm: 10000000
    otpm: 800000
  - tier: Tier 4
    model: Claude Sonnet 4.x
    rpm: 4000
    itpm: 2000000
    otpm: 400000
  - tier: Tier 4
    model: Claude Haiku 4.5
    rpm: 4000
    itpm: 4000000
    otpm: 800000
batchApiLimits:
  - tier: Tier 1
    rpm: 50
    queueMax: 100000
    perBatchMax: 100000
  - tier: Tier 2
    rpm: 1000
    queueMax: 200000
    perBatchMax: 100000
  - tier: Tier 3
    rpm: 2000
    queueMax: 300000
    perBatchMax: 100000
  - tier: Tier 4
    rpm: 4000
    queueMax: 500000
    perBatchMax: 100000
managedAgentLimits:
  - endpoint: Create endpoints (agents, sessions, environments)
    limit: 300
    window: minute
  - endpoint: Read endpoints (retrieve, list, stream)
    limit: 600
    window: minute
policies:
  - name: Cache-aware ITPM
    description: On most models, cache_read_input_tokens do NOT count toward ITPM, making prompt caching
      an effective way to increase throughput.
  - name: Auto tier advancement
    description: Tiers advance automatically based on cumulative credit purchase thresholds.
  - name: Acceleration limits
    description: Sharp usage spikes can trigger 429s independent of tier limits — ramp gradually.