DeepSeek, an AI company based in China, released an update to its V4.1 Flash AI model on Oct. 17. The new version, called V4.1-Flash-1017, is what DeepSeek calls a minor update, and the company says it scores higher on benchmarks, standard tests used to compare AI models, than the version it replaces, while finishing tasks faster and writing shorter answers. It is live on DeepSeek's website and app, and through its API, a connection that lets developers plug a service into their own apps. Anyone can download the model's files under the MIT license, a license that lets anyone use, change and share it free of charge.
DeepSeek says the update is tuned for coding tools, including its own DeepSeek Harness. DeepSeek says it trained the model with reinforcement learning, learning by trial and error, with a reward for getting it right. In its version, which it calls self-feedback, the aim was to get better at building whole projects from scratch. DeepSeek's examples of what the model built include a 3D model of Beijing's Forbidden City, a racing game that runs in a web browser and a version of the game Portal 2.
On DeepSWE, a coding test, DeepSeek's table gives the update 77.0, up from 74.2 for the version it replaces and ahead of Anthropic's Claude Opus 5 at 74.0 and OpenAI's GPT-5.6 Sol at 73.0. In the same table, Claude Opus 5 scores higher on 7 of the 11 tests, including Terminal-Bench 4.0, a coding test for AI agents, AI that can carry out tasks on its own, such as using apps and websites, instead of only answering questions. There, Opus 5 reaches 51.8 to DeepSeek's 42.3. The table leaves out newer models such as Anthropic's Claude Opus 5.5, released Sept. 22.
Through DeepSeek's API, the model costs $0.30 per million tokens of new text sent in, or $0.006 for text it has recently seen, and $1.20 per million it writes, at peak rates. Tokens are the small chunks of text, often parts of words, that AI models read and write. Rates are half that outside peak hours, which DeepSeek sets at 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays, except Chinese public holidays. DeepSeek's V4 Pro model costs $1.32 for new text sent in and $3.96 for text it writes at peak, and requests sent under the retired V4 Flash names now go to the new model.
DeepSeek also opened its own search engine to all developers, as two services that search the web and fetch pages, at $0.75 per 1,000 calls each. The company says the engine already handles hundreds of millions of requests a day. For DeepSeek API customers, searches and page fetches made through DeepSeek Harness and the coding tools Claude Code and Codex are free until billing starts on Oct. 31.
All of these scores come from DeepSeek's own table. It is not yet clear how the update compares with the newest models from Anthropic and OpenAI on tests run by outside groups.