Expired50% confidenceFactExact time
DeepSeek uses NVIDIA cards on V3 but not the official software libraries, writing operators at the kernel level using TileLang, with an efficiency loss of approximately 1% to 2%.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/25/2026
First Seen
This claim has expired. It is kept for the record and no longer describes the current state.
Valid until: 10/23/2026(expired)
Sources
梁文锋DeepSeek交流会：克制定价与AGI路线图解读
Related Claims
VerifiedDeepSeek-V3 outperformed open-source models such as Qwen2.5 and Llama 3.1 in multiple benchmarks.74% similarUnverifiedThe DeepSeek open platform offers two main levels of models: DeepSeek V3 (fast response speed, low token cost) and the DeepSeek R1 series (reasoning-enhanced version).73% similarUnverifiedThis release of DeepSeek V4 does not yet support multimodality and focuses solely on text capabilities.73% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/615392API
curl https://kongchang.com/api/v1/knowledge/claims/615392MCP
get_claim(id=615392)