Unverified50% confidenceFactExact time
Taking OpenAI o3 as an example, the number of internal reasoning tokens for a single complex request can be several times greater than the final output.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/6/2026
First Seen
Valid until: 1/4/2027
Sources
Related Entities
Related Claims
UnverifiedA single complex reasoning task with OpenAI's o1-pro can generate tens of thousands of hidden thinking tokens, and the cost may be 5–10 times that of normal mode.76% similarUnverifiedOpenAI的API采用Token计费机制，Token约等于3/4个英文单词或1-2个中文字68% similarUnverifiedOpenAI O系列模型的推理Token虽然不会出现在最终输出中，但仍然计入Token用量和费用68% similarUnverifiedThe OpenAI o1 series uses implicit CoT (thinking tokens), where the reasoning process is invisible to users but still consumes computational resources.67% similarUnverifiedMainstream large model APIs (such as OpenAI and Anthropic) charge based on the number of input tokens, and every character in a tool definition is a billable unit.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/980955API
curl https://kongchang.com/api/v1/knowledge/claims/980955MCP
get_claim(id=980955)