Unverified50% confidenceBenchmarkExact time
Using /compact between tasks cost $2.34, 17% more expensive than doing nothing, because compaction requires reading the entire context before compressing it.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/8/2026
First Seen
Valid until: 2/6/2027
Sources
Related Entities
Related Claims
UnverifiedAfter a long interruption (cache expiry), compaction is actually cheaper than simply continuing the conversation, because each subsequent turn then reads only about 15k tokens instead of the entire context.66% similarUnverifiedIn this cost breakdown, user input and attachments account for 21.6%, Shell command output 6.1%, subagents 2.1%, file reads 1.2%, search (grep/glob) 0.1%, MCP and other tools 0.1%, and web scraping 0.0%.63% similarVerifiedA 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.62% similarUnverified提升分词器压缩率意味着在相同 token 预算内可以容纳更多原始文本,直接降低 API 调用费用62% similarUnverifiedUnsloth abandoned the conversation summarization approach because the benefit of summarization was minimal yet added approximately 190 seconds of overhead per compression62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1147985API
curl https://kongchang.com/api/v1/knowledge/claims/1147985MCP
get_claim(id=1147985)