Unverified50% confidenceBenchmarkExact time
A calculation example: asking 20 questions about a 50,000-token document requires about 1 million input tokens without prompt caching, and about 157,000 with caching, reducing input cost by about 84%, with faster time to first token as well.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/16/2026
First Seen
Valid until: 1/14/2027
Sources
Related Entities
Related Claims
UnverifiedAccording to community benchmarks, in complex sessions where the context has not been cleared, a single task investigation can consume up to 500,000 tokens, generating 4–6 times the overhead compared to short sessions used out of the box.73% similarUnverifiedIn 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%).71% similarUnverifiedOverly long system prompts bring hidden overhead: for example, when the system prompt has 500 tokens while the actual input has only 50 tokens, it means 91% of the cost goes to boilerplate text; auditing and trimming the system prompt by 40% can reduce costs.71% similarUnverified在1.78百万token的长会话实测中，因为97%的token被缓存，发送token最多的配置反而最便宜70% similarUnverifiedThe first turn after warm compaction costs about $0.026–$0.032, even smaller than a brand-new session, because the skill manifest is no longer sent and files over 5,000 tokens are returned only as path references.69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1048064API
curl https://kongchang.com/api/v1/knowledge/claims/1048064MCP
get_claim(id=1048064)