SAN FRANCISCO, Oct 19 — Chinese artificial intelligence start-up DeepSeek has released a faster, leaner version of its AI model, reporting improved benchmark scores and shorter response times compared with its predecessor.
Bloomberg reported that the Chinese AI developer released DeepSeek-V4.1-Flash-1017 on October 17, with the updated version available through its website, app and developer platform.
In results published on Hugging Face, DeepSeek said the new model scored 42.6 on Humanity’s Last Exam without tools, up from 36.8 for the previous V4.1-Flash version.
The benchmark tests AI systems on difficult questions spanning mathematics, science and the humanities, while DeepSeek’s technical report also listed improvements in software development and other tasks involving AI agents.
The company said the update was designed to reduce the time needed to complete tasks and produce shorter responses, potentially lowering the cost of using the model.
Its reported results also showed an increase on Terminal-Bench 4.0, which assesses coding agents, from 31.2 to 42.3, while its DeepSWE v1.1 score rose from 74.2 to 77.0.
These figures are based on DeepSeek’s own published evaluations and do not, on their own, establish how the model performs across all real-world uses.
The release comes as Chinese AI developers compete with US firms such as OpenAI and Anthropic, as well as local rivals, to deliver increasingly capable systems at lower costs.
DeepSeek’s September launch of the smaller V4.1-Flash model had put pressure on competitors, with shares of MiniMax Group, Z.ai and Alibaba Group Holding falling on the Hong Kong stock market after the earlier release, according to Bloomberg.
The latest update follows recurring service slowdowns and an outage lasting about five hours earlier in the week, which left DeepSeek’s servers inaccessible.
Some experts attributed the disruption to a shortage of computing resources, as demand for the company’s competitively priced AI models has put pressure on its services.
