Unverified50% confidenceFactExact time
Standard reasoning mode does not use long chain-of-thought, and token consumption is typically one-tenth or even less than that of high reasoning mode.
1
Sources
50%
Confidence
Long-term
Relevance
7/24/2026
First Seen
Sources
Codex超高推理模式隐藏成本：一个任务耗尽70%额度的真相
redditr/OpenAI7/12/2026
Related Claims
Unverified将概念推理量化为单一指数可能丢失多维度、情境依赖的推理能力信息，误导下游应用选型决策73% similarUnverifiedThe author recommends disabling thinking, on the grounds that most use cases for building agents or simple applications do not require deep reasoning, and LangChain's compatibility with thinking mode is not yet complete.73% similarUnverified单一基准的高分不等于全面领先，Terminal-Bench无法反映模型在长文本理解、多模态、复杂推理等维度的表现71% similarUnverifiedUltra模式与提高推理强度有本质区别，高推理强度是让单个智能体花更多时间思考70% similarUnverifiedAnthropic的extended thinking模式下的内部推理token会单独计费68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/608403API
curl https://kongchang.com/api/v1/knowledge/claims/608403MCP
get_claim(id=608403)