Unverified60% confidenceFactExact time
DeepSeek自研了高级语言TileLang，用它重写CUDA算子，代码量更小、效率更高，V3训练已基本不依赖英伟达生态
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
7/25/2026
First Seen
Valid until: 10/23/2026
Sources
Related Claims
UnverifiedDeepSeek-V3以极低的训练成本实现了顶尖推理能力72% similarUnverifiedDeepSeek以极低的训练成本实现了与顶级闭源模型接近的数学与代码能力，其R1模型在发布时引发全球关注71% similarUnverifiedDeepSeek V3在预训练阶段引入了FP8混合精度训练框架，降低了训练显存占用与通信开销71% similarUnverifiedDeepSeek V4后续版本会支持原生多模态，但多模态只是一个组件，主线仍是把AI训练做好并解决持续学习71% similarVerifiedDeepSeek V4在编程、数学推理和多语言理解方面表现突出71% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/613786API
curl https://kongchang.com/api/v1/knowledge/claims/613786MCP
get_claim(id=613786)