Unverified50% confidenceFactExact time
Thinking tokens are the tokens a reasoning model consumes for internal reasoning before producing an answer; they are usually not shown directly to the user but are still billed as output tokens.
1
Sources
50%
Confidence
Long-term
Relevance
10/18/2026
First Seen
Sources
Related Entities
Related Claims
Unverified推理 Token（Reasoning Token）概念源自 OpenAI o 系列模型，模型在输出可见答案前会先消耗 Token 进行内部推导，这些中间步骤对用户不可见但消耗真实算力78% similarUnverifiedReasoning mode performs internal chain-of-thought reasoning before the final answer, and the thinking tokens generated are billed; in complex programming tasks, thinking tokens may account for 60%–70% of total consumption, and costs may be 3–5 times that of normal mode.77% similarUnverified对于大语言模型推理，吞吐通常以 token 数为单位衡量，即 tokens/s 是评估 LLM 推理框架性能的核心指标75% similarUnverifiedThe reasoning tokens produced by reasoning models share the same max_tokens budget as the final output, and they are billed as usual even if the visible reply is an empty string.71% similarUnverifiedWhen the model is fixed, the tool shapes token consumption, not output quality.70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1059531API
curl https://kongchang.com/api/v1/knowledge/claims/1059531MCP
get_claim(id=1059531)