Unverified50% confidenceFactExact time

DeepSeek-R1等推理增强模型在生成最终答案前会产生大量中间推理Token，这些思考Token计入输出Token费用

1

Sources

50%

Confidence

Long-term

Relevance

8/30/2026

First Seen

Sources

DeepSeek API涨价3倍：还能斩杀所有Coding Plan吗？

Related Entities

DeepSeek-R1model 思维链concept

Related Claims

UnverifiedDeepSeek-R1 用特殊标签（如 <think>...</think>）在输出中显式区分思考过程与最终答案79% similar UnverifiedDeepSeek-R1的发布引发业界对推理时间计算（inference-time compute）范式的广泛讨论75% similar UnverifiedDeepSeek V3和R1借助MoE架构在保持顶级性能的同时将推理成本压缩至传统Dense模型的数分之一74% similar

Cite This Claim

Stable URI

https://kongchang.com/claim/825677

API

curl https://kongchang.com/api/v1/knowledge/claims/825677

MCP

get_claim(id=825677)