On September 10, 2026, DeepSeek officially released V4.1 Flash, a 552-billion-parameter model with approximately 8 billion activated parameters for input and 16 billion for decoding. The model supports a one-million-token context window, image input and is open-sourced under the MIT license.
Internal and external testing showed V4.1 Flash surpasses the previous V4 Pro in performance, response speed, computational cost and total processing time. DeepSeek also cut Flash series pricing by up to 60%, with cache-hit input dropping to 0.02 RMB per million tokens.
The company said V4 Pro will be automatically routed to V4.1 Flash as the default backend starting September 14. Tencent WorkBuddy and CodeBuddy have fully integrated the new model.
The release further strengthens DeepSeek's position in the open-source large model market and lowers application barriers for developers.
Source: DeepSeek / Tencent News
