Unverified50% confidenceDecision RuleExact time
When context usage reaches 70%–90%, it is best to compress the context or start a new conversation, because model performance degrades as it approaches the end of the context window.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/16/2026
First Seen
Valid until: 12/15/2026
Sources
Related Entities
Related Claims
Unverified当对话轮次过多超过Token上限时，需采用滑动窗口、摘要压缩、关键信息提取等策略管理上下文长度77% similarUnverifiedAs the number of multi-turn conversation rounds increases, the context window becomes occupied by preceding dialogue, the signal strength of citation instructions in the prompt relatively weakens, and citation quality may undergo systematic degradation.72% similarUnverifiedCodex's context compression mechanism can condense a conversation into a summary of key conclusions; in testing it reduced usage from 92% to 3%, and can also be triggered manually with the slash command '/compress'.72% similarUnverifiedWhen the context window is nearly full, context compression is performed by discarding irrelevant information or generating summaries, in order to control token costs and keep the model's attention focused.71% similarUnverifiedAn overly long CLAUDE.md compresses the context window space available for actual code and conversation, reducing the model's effective working capacity.71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/924591API
curl https://kongchang.com/api/v1/knowledge/claims/924591MCP
get_claim(id=924591)