PRODUCTION SCENARIO
A 2,000-token system prompt carrying a cache breakpoint reports cache reads on Claude Sonnet 5. The identical prompt on Claude Haiku 4.5 reports zero cache creation and zero cache reads, and the API returns no error.
What explains the difference between the two models?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: Claude Haiku 4.5 has a higher minimum cacheable prefix length than Sonnet 5
The minimum cacheable prompt length is set per model: 1,024 tokens on Claude Sonnet 5 but 4,096 on Claude Haiku 4.5, so a 2,000-token prefix caches on one and silently does not on the other, with no error returned. Rate-limit accounting is a separate concern and does not change what the usage fields report. The context window governs how much fits in a request, not what is eligible to cache. And ttl is optional on every model; omitting it selects the 5-minute default rather than disabling caching.