Unverified75% confidenceOpinionTime unknown
当前推理模型在数学和编程领域表现突出，但在开放域推理、因果推理等方面仍有提升空间
1
Sources
75%
Confidence
Medium-term (~90 days)
Relevance
6/1/2026
First Seen
Valid until: 8/30/2026
Sources
LLM推理能力演进：从思维链到DeepSeek-R1全解析
githubatfortes
Related Entities
Related Claims
Unverified推理增强模型本质上是以推理时算力（Inference-time Compute）换取准确率73% similarUnverified深度推理模式的技术基础是链式思维（Chain-of-Thought）推理与推理时计算扩展（Inference-time Compute Scaling）73% similarVerified推理时扩展（inference-time scaling）通过在推理阶段增加计算量提升模型性能，存在类似预训练的幂律关系73% similarUnverified推理时计算扩展通过延长思维链、多路径采样择优、自我验证反思等方式在推理阶段投入更多计算可提升复杂推理任务性能72% similarUnverifiedAnthropic的Extended Thinking（扩展思考）功能允许模型进行更长、更深入的推理，是Opus模型在模糊场景中表现出色的关键原因72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/29117API
curl https://kongchang.com/api/v1/knowledge/claims/29117MCP
get_claim(id=29117)