DeepSeek-V4-Pro
DeepSeek class A live
DeepSeek's flagship 1.6T-param MoE (~49B active), 1M context, MIT open weights, with thinking + non-thinking modes.
Layer 1 — factual axes
Layer 2 — output quality
44 / 100
DeepSeek’s flagship 1.6T-param MoE (~49B active), 1M context, MIT open weights, with thinking + non-thinking modes.