DeepSeek releases V4.1-Flash-1017, claims 77.0 on DeepSWE
Chinese AI company DeepSeek has rolled out V4.1-Flash-1017, a new checkpoint of its V4.1-Flash model, on its website, app and API, alongside a search API.
DeepSeek, the Chinese artificial intelligence company, released DeepSeek-V4.1-Flash-1017 on 17 October 2026. The model is a new checkpoint of the company's existing V4.1-Flash and is available immediately through DeepSeek's website, its app and its API.
DeepSeek says the update improves on the V4.1-Flash version it shipped in September on several fronts. According to the company, the new checkpoint posts stronger benchmark scores, finishes tasks faster and produces shorter outputs. The 1017 suffix appears to refer to the release date.
Benchmark claims
The model is reported to score 77.0 on DeepSWE, a benchmark focused on software engineering tasks. That figure comes from DeepSeek itself, and no independent evaluation has yet been published to confirm it. The full set of comparative results against the September release was not available in the reporting reviewed.
The company has also opened a search API to developers as part of the launch. Details on its pricing, coverage and technical design were not included in the available information.
Why it matters
Models branded Flash are typically aimed at speed and lower cost rather than maximum capability. Shorter outputs and quicker task completion, if they hold up in practice, would cut token usage and latency for developers running high volumes of requests through the API.
DeepSeek has become one of the most closely watched Chinese AI developers, known for competitive models offered at low prices. Shipping a refreshed checkpoint roughly a month after the September version suggests the company is keeping up a rapid update cadence. How the new model performs outside DeepSeek's own testing will become clearer as independent benchmarks appear.
Comments
No comments yet. Start the discussion or ask a question.
Related
Chinese AI company StepFun has published open weights for Step 5 Preview, a 600-billion-parameter sparse model with 27 billion active parameters and a 1 million token context.
Vercel data shows TypeSafe AI's decision model Jev reached 35% of paid teams within two weeks of launch, nearly double the reach of runner-up Sonnet 5.5.
The Granite team released Granite Switch 4.1 on October 15, 2026, combining chat, decision-making adapters and speech-to-text in a single hosted endpoint.