Concept保留式推理
Preserved Thinking
A reasoning-state persistence mechanism developed by Anthropic for Claude models. During long sessions involving sequential tool calls, it retains previous analytical reasoning states in the form of an active cache, improving coherence and efficiency across multi-step tasks.
Timeline (last 90 days)
Oct 11
The Preserved Thinking mechanism avoids KV Cache invalidation and reasoning state loss caused by tool calls by fixing previous reasoning chains in a structured form within the context prefix region.
Unverified50%