product · May 14, 2026
DeepSeek Launches V4 AI Model
Share the canonical public link.
DeepSeek released its V4 AI model on May 14, 2026, featuring benchmark scores competitive with U.S. counterparts. The model supports advanced reasoning and agentic tasks, trained on over 30 trillion tokens using under $6 million in compute for prior versions. Lian Jye Su of Omdia noted V4's edge in efficiency over GPT-5.5 and Claude 4. DeepSeek's V3 predecessor achieved 95.83 percent on AIME 2026 math benchmarks, surpassing OpenAI's o3 by 5 points.
Failed after 3 attempts. Last error: Service unavailable