Unverified50% confidenceFactExact time
The author has a spending dashboard that can show that 63% of the context is bash output and that a single web search consumes 27,000 characters, but it never showed that one third of the spending went to buying back one's own mistakes.
1
Sources
50%
Confidence
Long-term
Relevance
10/15/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedAccording to the author's logs, actual input content accounted for only about 0.01% of the bill, with 96% of the cost coming from old conversation history being resent repeatedly.68% similarUnverifiedIn the session assumed by the article, a gateway labeled "50% off" but with cache reads at only a 20% discount and cache writes at full price actually gives a 28% discount, costing $12.13.64% similarUnverifiedA Prompt Caching cache hit read costs only 0.1 times the base input price, while the initial write cost is 2 times the regular input price.63% similarUnverifiedThe script did not enable prompt caching, causing all 375 million input tokens to be billed at the uncached price.63% 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.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1033123API
curl https://kongchang.com/api/v1/knowledge/claims/1033123MCP
get_claim(id=1033123)