Unverified60% confidenceFactExact time
DeepMind 研究表明在某些数学推理任务上通过增加推理时计算量，较小模型可以超越参数量更大的模型
2
Sources
60%
Confidence
Long-term
Relevance
7/6/2026
First Seen
Sources
Claude Sonnet 5真的便宜吗？按任务重新算账
bilibili为什么叫QQ7/1/2026
Related Claims
UnverifiedThinking Level in DeepSeek models is a capability that allows callers to control how much reasoning the model performs before generating a response, representing a practical application of Test-Time Compute Scaling.77% similarUnverified在同等参数规模下 DeepSeek Math 的数学推理能力强于 Llama、Qwen 和 WizardMath 等开源模型75% similarUnverifiedDeepSeek-R1等模型被部分研究者视为在算力受限条件下通过算法创新实现的「降维突围」，在参数效率和推理能力上逼近国际前沿73% similarUnverifiedDeepSeek模型体系中deepseek-chat是非推理模型，deepseek-reasoner是带思考过程的推理模型72% similarUnverifiedDeepMind、Google Research、Meta AI等机构青睐有强数学背景的人才，尤其在理论机器学习、优化算法、生成模型的数学分析等方向72% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/113288API
curl https://kongchang.com/api/v1/knowledge/claims/113288MCP
get_claim(id=113288)