Unverified50% confidenceFactExact time
In this audit, 85% of cache write tokens used 1-hour prompt caching, whose write unit price is higher than 5-minute caching; but reverse testing showed that switching everything to 5-minute caching would actually cost $1,781 more (+22.6%).
1
Sources
50%
Confidence
Long-term
Relevance
10/16/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedOn most Claude models, Anthropic bills cache reads at 0.1× the base input price; default 5-minute cache writes at 1.25×; 1-hour cache writes at 2×; and a 5-minute cache entry is refreshed for free each time it is used.80% similarUnverifiedIn this audit sample, 66% of the cost came from cache reads, 99% of input-side tokens were cache hits, while output (including thinking) accounted for only 12%.74% similarVerifiedThe cache read mechanism retains the KV cache for recurring long prompt prefixes, charging at an extremely low rate upon a cache hit; the leaked cache read price for Opus 5.5 is approximately 1/20 of the input price.74% similarUnverifiedCached input tokens typically enjoy a discount of 25%–50% off the original price, as there is no need to recompute embeddings and attention weights—only the precomputed KV vectors need to be loaded from cache.69% similarUnverifiedThe prompt caching mechanism allows already-processed input prefixes to be cached server-side, and if the next request starts with the same prefix, it is read at a price far below normal input; common cache TTLs are 5 minutes and 1 hour, and writing to the cache costs extra.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1047788API
curl https://kongchang.com/api/v1/knowledge/claims/1047788MCP
get_claim(id=1047788)