Chinese artificial intelligence startup DeepSeek released DeepSeek-V4.1-Flash-1017, an updated model it says completes tasks faster, and opened its search engine to external developers.
The updated model is available in DeepSeek’s web and mobile applications. API customers can access it through the existing deepseek-flash model ID at unchanged prices.
DeepSeek’s model card on Hugging Face calls the release a “minor update” that “reduces task execution time and output length.” The company did not quantify either reduction.
The release comes a little more than five weeks after Hangzhou DeepSeek Artificial Intelligence Basic Technology Research Co. Ltd. launched the original V4.1-Flash as the smallest model in a new architecture family.
In DeepSeek’s testing, the update resolved 77% of tasks on the DeepSWE v1.1 software engineering benchmark, compared with 74.2% for the September version. Anthropic PBC’s Claude Opus 5 scored 74% in the same table.
Anthropic’s model led on seven of the 11 benchmarks published by DeepSeek. Its lead on Humanity’s Last Exam was nearly 14 points.
DeepSeek said it optimized the update for agent software, including its DeepSeek Harness tool and Tencent Holdings Ltd.’s CodeBuddy coding assistant and WorkBuddy desktop agent for office work. WorkBuddy has offered V4.1-Flash since the original model’s release on Sept. 10.
DeepSeek’s new Search API gives external developers access to its in-house search engine and returns up to 50 ranked web results per query. A separate Fetch API converts a single web page into Markdown.
Both APIs cost 75 cents per 1,000 successful requests. Calls through DeepSeek Harness, Anthropic’s Claude Code or OpenAI Group PBC’s Codex are free until Nov. 1.
At the original model’s September launch, off-peak API pricing started at 15 cents per million uncached input tokens. Bloomberg reported that shares of Chinese rivals MiniMax Group Inc. and Alibaba Group Holding Ltd. fell in Hong Kong after that release.
The latest releases follow an API outage on Oct. 15 that lasted nearly five hours. Yicai Global counted it among DeepSeek’s worst outages of the year, while some experts quoted by Bloomberg attributed it to a shortage of computing capacity.
DeepSeek apologized for the outage and credited users based on their spending during the previous week. The company said it would remove single points of dependency in its systems to prevent a recurrence.
DeepSeek’s change log recommends Harness version 0.2.1-rc.1 or later for the updated model. The model weights are available on Hugging Face under the MIT license.
