PRODUCTION SCENARIO
A developer models the cost of an interactive code analysis tool on Claude Sonnet 5. A 10,000-token repository context is sent across 50 consecutive queries in one session with prompt caching enabled.
How is a cache read priced against base input tokens on this model?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: Cache reads are billed at 10% of the base input token rate, a 90% saving
Cache read tokens are billed at 0.1x the base input price on Claude Sonnet 5 and on most models, so 49 of the 50 requests pay a tenth for the repository context. The 1.25x figure is the 5-minute cache write multiplier, paid once when the entry is created, not on reads. Caching is not a latency-only optimization, and reads are billed rather than free. Claude Fable 5.1 and Mythos 5.1 are the documented exception, pricing reads at 0.025x.