Unverified50% confidenceBenchmarkExact time
DeepSeek V3, released on December 26, 2024, has 671 billion total parameters and 37 billion activated parameters, uses FP8 training, reports a compute cost of $5.576 million in its paper, and matches GPT-4o in performance.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/23/2026
First Seen
Valid until: 1/21/2027
Sources
Understanding DeepSeek in 3 Minutes: The Rise from a Quant Hedge Fund to an Open-Source AI Leader
bilibiliGerrardYX10/23/2026
Related Entities
Related Claims
Unverified训练DeepSeek-V3仅消耗约278万H800 GPU小时，成本约合550万美元72% similarUnverifiedIn early 2024, DeepSeek-R1 reportedly reproduced reasoning capabilities close to GPT-4o at a training cost of less than $6 million, triggering a single-day evaporation of hundreds of billions of dollars in global AI stock market value.71% similarUnverifiedDeepSeek published a paper in the journal Nature, putting R1's final training cost at $294,000 using 512 export-compliant NVIDIA chips.70% similarUnverifiedA developer following Karpathy's nanochat project trained 9 checkpoints across 3 different versions of LLMs from scratch with a budget of approximately $750, with parameter counts ranging from 353M to 672M.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1090498API
curl https://kongchang.com/api/v1/knowledge/claims/1090498MCP
get_claim(id=1090498)