DeepSeek ships V4.1-Flash-1017 as open weights and opens its Search API

DeepSeek has released V4.1-Flash-1017, live on its site, app, and API at unchanged pricing, with the model weights published under an MIT license on Hugging Face and ModelScope. DeepSeek reports gains on all 11 of its published benchmarks, including 77.0 on DeepSWE v1.1, while Claude Opus 5 still leads on raw reasoning measures such as HLE. The company has also opened its Search and Fetch APIs to the public at $0.75 per 1,000 calls.

Independent benchmarks of the new checkpoint have not yet been published; the figures above are DeepSeek's own.

Sources: The Decoder; MarkTechPost