Cache-aware ITPM
Catalog row in Vendor License Rules · Cited
- Kind
- Counting
- Statement
- Rate limits (as published on 2026-09-26): "For most Claude models, only uncached input tokens count toward your ITPM rate limits." input_tokens and cache_creation_input_tokens count toward ITPM; cache_read_input_tokens do not for most models (Claude Haiku 3.5 is the listed exception). OTPM counts actual tokens generated; max_tokens does not factor in.
- Applies when
- Estimating throughput headroom when prompt caching is used.
- Applies to
- Claude API Messages endpoint.