DeepSeek-V4.1-Flash-1017: MIT open-weights update, stronger on agentic tasks and cheaper per task

In one sentence On October 17 DeepSeek published V4.1-Flash-1017, a revision of its Flash model with MIT weights on Hugging Face and ModelScope, a 1M-token context and gains on coding and agent tasks. It also opened its Search and Fetch APIs to everyone at 0.75 dollars per 1,000 calls.

Verified Official source

ShareLinkedIn X

DeepSeek released a new version of its fast model, V4.1-Flash, tagged with the date: 1017. It is not a new model from scratch but a revision: according to the company it answers better, finishes tasks faster and writes shorter replies, so the same work costs less.

The weights can be downloaded from Hugging Face and ModelScope under the MIT license, the most permissive one: anyone with suitable hardware can run it locally and use it commercially. The model is also available on the website, in the app and through the API.

Alongside the model DeepSeek opened its web search APIs (Search and Fetch) to all developers, built on the company's in-house search engine. Until October 31, calls made through server-side tools in DeepSeek Harness, Claude Code and Codex are free.