DeepSeek V4.1-Flash-1017 launched across web, app, and API, with optimizations for DeepSeek Harness, Tencent CodeBuddy, and WorkBuddy.

DeepSeek updated its existing V4.1 Flash model to V4.1-Flash-1017. According to the company announcement relayed by Financial_Express, the official website, app, and application programming interface, or API, were all switched to the new model.

The reported training change centers on self-feedback-based reinforcement learning. DeepSeek says this improved the model’s ability to construct complete projects from scratch, including game development and three-dimensional modeling. The announcement presents these improvements as supporting more natural-language programming requests and project-development tasks.

The release also targets particular development platforms and frameworks. DeepSeek identifies optimizations for DeepSeek Harness, Tencent CodeBuddy, and WorkBuddy and claims substantial benchmark improvements over the preceding V4.1 Flash version. The supplied messages do not include benchmark scores, evaluation settings, pricing changes, or model-size details.