DeepSeek released DeepSeek-V4.1-Flash-1017 on October 17, 2026, an update to the V4.1-Flash model first released in September. DeepSeek describes it as delivering “Stronger benchmark performance, faster task completion, and shorter outputs than the previous version,” which it calls a Pareto improvement in performance and cost. It went live the same day on DeepSeek’s website, app and API, with weights published on Hugging Face and ModelScope.
The update was trained with reinforcement learning using self-feedback to improve end-to-end project development, which DeepSeek illustrates with generated 3D scenes and games. It is optimized for DeepSeek Harness, Tencent CodeBuddy and WorkBuddy, and DeepSeek recommends Harness v0.2.1-rc.1 or later. DeepSeek’s API docs say the deepseek-flash model name is now served by V4.1-Flash-1017, and that requests to the retired names deepseek-v4-flash and deepseek-v4-flash-vision-exp are also routed to it at the Flash price.
In the same announcement DeepSeek opened its Search and Fetch APIs to all developers, powered by its in-house search engine, priced at $0.75 per 1,000 calls per endpoint. Calls made through server-side tools in DeepSeek Harness, Claude Code and Codex were free for DeepSeek API users, with billing starting October 31, 2026 at 16:00 UTC.
Why it matters: DeepSeek is now shipping a weights-open model together with its own agent harness and a first-party web search API, giving it a full agent stack rather than just a model, and its explicit support for running inside Claude Code and Codex lowers the cost of swapping it in under rival tools.
What it does not show: the announcement presents performance as a chart against cost on DeepSWE v1.1 measured by DeepSeek with its own harness, with no independent evaluation, and it does not restate parameter counts or a license for the updated checkpoint.