Unverified60% confidenceFactExact time
Claude的扩展思考（Extended Thinking）模式会让模型在输出前进行显式内部推理链，提升复杂任务质量，但Token消耗量大幅增加
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
7/20/2026
First Seen
Valid until: 10/18/2026
Sources
Related Claims
VerifiedClaude的Extended Thinking功能开启后，模型会在生成最终回答前先输出一段内部推理过程（thinking块），这段推理对用户可见但不计入对话上下文82% similarUnverified扩展思考模式会让模型在输出前进行显式内部推理链，代价是Token消耗大幅增加且推理Token通常按更高费率计费81% similarUnverifiedReasoning Effort本质上是控制模型思维链深度的参数，档位越高模型允许的内部推理步骤越多，延迟和成本也成比例上升80% similarUnverified扩展思维（Extended Thinking）机制的核心是在模型的推理阶段分配更多的计算预算，让模型能够对问题进行多角度分析、自我纠错和方案比较，学术界称之为推理时计算缩放（Inference-Time Compute Scaling）78% similarVerifiedExtended Thinking is a reasoning enhancement mechanism introduced by Anthropic in the Claude 3 series that allows the model to perform an explicit internal Chain-of-Thought before generating a final answer.78% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/566530API
curl https://kongchang.com/api/v1/knowledge/claims/566530MCP
get_claim(id=566530)