DeepSeek releases V4.1-Flash-1017: open weights under MIT and a coding score above Claude Opus 5 on one benchmark
Prepared by Sintelligent with the help of SI. Published in SI: . Updated: .
A strong model for coding and SI agents can now be downloaded and run by anyone, and tasks on it, according to its authors, cost less.
The model's weights are published on Hugging Face under the MIT license, and it can be tried on DeepSeek's website, in its app and through its API. According to DeepSeek's own results, on the DeepSWE v1.1 test the new version scored 77 against 74.2 for the previous V4.1-Flash, ahead of Claude Opus 5 (74) and GPT-5.6-Sol (73). Opus 5 leads on Terminal-Bench 4.0, ProgramBench, NL2Repo-Bench and HLE without tools (on HLE, 56.3 against 42.6). By The Decoder's estimate, a DeepSWE task costs about $0.20 versus about $0.50 for the previous version. At the same time DeepSeek opened a Search API and a Fetch API built on its own search engine at $0.75 per 1,000 successful requests; according to The Decoder, they are free until October 31 for users of Claude Code, DeepSeek Harness and Codex.