DeepSeek has formally released its V4.1 Flash model and updates its API routing so that V4 Pro requests are sent to Flash. According to notifications to API users, DeepSeek says V4.1 Flash performs better than V4 Pro in internal and external testing, including measures such as cost, speed, and total completion time. The change is scheduled to begin on Sept. 14 at noon Beijing time.
Reporting across outlets describes V4.1 Flash as a smaller model positioned as a new default for coding and agent-style tasks, with pricing aligned to the Flash series. One outlet notes that DeepSeek’s Hangzhou lab announces the release publicly and that the model weights are published on Hugging Face under the MIT license, enabling download, modification, and local use. Another outlet focuses on the operational update for existing API users—specifically the transition away from direct V4 Pro handling and the introduction of new Flash-series pricing.
Overall, coverage agrees on the launch of V4.1 Flash, the claim of improved performance versus V4 Pro, and the start of new request-routing and pricing changes, while emphasizing either model availability or API operational details.
