According to CoinMeta, DeepSeek announced the release of the official version of V4-Flash. However, V4-Pro still needs to wait for the launch of that version's API. The new version focuses on improving programming capabilities, tool invocation, and the execution of long tasks. Officials claim that several agent benchmark scores have significantly surpassed those of previous versions. The new version natively supports responses and API, and has been adapted for Codex. There have been no changes to the model structure or parameter scale; the main improvements were made in post-training. This update only affects the API interface, with no updates to DeepSeek App or the web interface. DeepSeek also released the official version of V4-Flash-0731 and continues to use the MIT license. Developers can deploy, fine-tune, and use it commercially on their own, without being restricted to using it only through DeepSeek's official API.