DeepSeek releases faster V4.1-Flash update and opens search tools to developers
The Chinese AI company says the new version scores higher on benchmarks while finishing tasks faster with shorter outputs.
Chinese artificial intelligence company DeepSeek released an updated model, DeepSeek-V4.1-Flash-1017, on Saturday, saying it delivers “stronger benchmark performance, faster task completion, and shorter outputs than the previous version.”
The Hangzhou-based company described the release as a minor update to V4.1-Flash, which it launched in early September, Bloomberg News reported. The new version is available on DeepSeek’s website, app and developer platform, and its weights have been published for download on Hugging Face and ModelScope.
In a technical report on Hugging Face, DeepSeek said the model brings “substantial improvements across reasoning, coding, agentic tasks, and visual benchmarks.” The company said it used reinforcement learning to improve the model’s ability to build complete projects, from games to 3D models.
On the Humanity’s Last Exam benchmark, the updated model scored between 36.8 and 42.6, compared with 36.8 to 39.1 for the earlier V4.1-Flash, according to figures DeepSeek posted on X, Bloomberg reported.
DeepSeek also opened its Search and Fetch application programming interfaces to all developers, priced at $US0.75 per 1,000 calls for each. Calls made through certain coding tools are free until billing starts on 31 October.
Earlier in the week, DeepSeek suffered an outage of about five hours that left its servers unreachable, Bloomberg said.
Sources
- DeepSeek-V4.1-Flash-1017: Wish-Driven Programming Made Real · DeepSeek
- DeepSeek Pushes Out Updated AI Model It Says Is Faster, Leaner · Bloomberg (via Yahoo Finance)
- deepseek-ai/DeepSeek-V4.1-Flash-1017 · Hugging Face