← Back
24 April 2026Open source
DeepSeek releases V4 preview with 1M-token context
DeepSeek open-sourced V4-Pro (1.6T parameters, 49B active) and V4-Flash (284B), using a new hybrid compressed attention that sharply cuts long-context compute and KV cache.
Actors involved
Tags
deepseek-v4open-weightslong-context
Sources
Hugging Facedeepseek-ai/DeepSeek-V4-Pro