Unverified60% confidenceFactExact time
推理模型的思考预算机制设有推理Token数量上限，耗尽后无法输出完整答案，可能导致过度思考问题
2
Sources
60%
Confidence
Long-term
Relevance
7/9/2026
First Seen
Sources
16G显存实测：Ornith 35B vs Qwen 35B全方位对决
bilibili程序员-智能译站7/7/2026
Related Claims
Unverified当模型规模跨越某些阈值后会涌现出此前不具备的能力（涌现能力），如多步逻辑推理和复杂数学计算74% similarUnverified强制推理模型输出结构化格式会干扰其内部推理过程并降低性能73% similarUnverified高强度推理配置通常意味着更高费用和更长响应时间，牺牲响应速度换取逻辑严密性72% similarUnverifiedQwen3 27B默认会对简单问题也启动完整长链推理流程，即存在过度思考（overthinking）问题72% similarUnverifiedReasoning Effort档位越高，模型可消耗的思考Token上限越大，但延迟和Token成本相应线性乃至超线性增长71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/342899API
curl https://kongchang.com/api/v1/knowledge/claims/342899MCP
get_claim(id=342899)