DeepSeek · model release · downloadable model
DeepSeek-V4 Preview
DeepSeek previews V4, a new generation of open models that can read very long documents.
In detail
DeepSeek released the DeepSeek-V4 Preview with two models, V4-Pro (1.6T total, 49B active parameters) and V4-Flash (284B total, 13B active), both with 1M-token context, open weights, on chat.deepseek.com and the API.
Sources
- DeepSeek-V4 Preview Release · DeepSeek · Primary source
Last checked 11 November 2026.