Source: Anthropic — Service tiers (Priority Tier)

Anthropic's legacy service-tier page: Standard, Batch and Priority Tier; Priority commitments in ITPM/OTPM with cache-weighted burndown, 99.5% uptime target, `service_tier` auto/standard_only. No longer sold as of late 2026.

Contents

Summary

Describes how the Claude API assigns a request to Priority, Standard or Batch capacity. As read on 2026-11-14 the page is marked legacy: Priority Tier commitments are no longer sold, existing ones run to contract end, and guaranteed capacity goes through sales.

Key claims

  • Standard is best-effort and the default. Priority Tier requests are served ahead of all others, which reduces 529 overloaded errors at peak.
  • A Priority commitment specifies input TPM, output TPM, a term (1, 3, 6 or 12 months) and one model version; it targets 99.5% uptime, and overflow falls back to Standard automatically.
  • Burndown against Priority capacity follows relative price: cache reads 0.1, 5-minute cache writes 1.25, 1-hour cache writes 2.0, US-only inference 1.1 on Claude 4.6+; other tokens 1.
  • Priority requests also count against regular rate limits; a request that would exceed them is declined.
  • service_tier: auto (default, use Priority if available) or standard_only. The response usage.service_tier reports the assigned tier; anthropic-priority-{input,output}-tokens-{limit,remaining,reset} headers show commitment state.
  • Priority Tier does not cover most 5.x-generation models (Opus 5/5.5, Sonnet 5/5.5, Haiku 5.5, Fable 5.1/5.2, Mythos 5.x and Mythos Preview); Claude Fable 5 is still supported, alongside the 4.5–4.8 models.

Feeds

Provisioned throughput, Rate limits and retries

Suggest changes →