Unverified70% confidenceFactExact time
Reasoning-enhanced models (such as o1 and DeepSeek-R1) consume large numbers of reasoning tokens in their chain-of-thought process that are invisible to users; the actual tokens consumed by o1-preview are often 3 to 8 times that of the visible output.
2
Sources
70%
Confidence
Long-term
Relevance
7/18/2026
First Seen
Sources
Related Claims
Unverified推理模型的中间思维链步骤在计费层面同样计入token消耗，OpenAI的o1模型思考token有时高达正常输出的数倍77% similarUnverifiedo1、DeepSeek-R1等模型将思维链内化为训练目标，在输出答案前生成大量内部草稿推理76% similarUnverifiedOpenAI o1、o3以及Anthropic Claude 3.7 Sonnet等推理模型会在返回答案前进行大量内部反思，中间推理Token对用户不可见但仍计入计费75% similarUnverifiedReasoning models perform extended reasoning within thinking blocks wrapped in special tokens (such as <think> and </think>); this content is typically not shown directly to users but consumes a large number of tokens.74% similarUnverified推理增强型模型如o1、DeepSeek-R1在API计费上采用输入+输出+推理token三段式结构74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/548619API
curl https://kongchang.com/api/v1/knowledge/claims/548619MCP
get_claim(id=548619)