DeepSeek released its V4.1-Flash-1017 model on Oct. 17, sharpening competition between Chinese open-weight AI models and higher-cost proprietary U.S. systems. The Hangzhou-based startup says the update improves performance while keeping costs low. The release extends a contest over who can deliver capable artificial intelligence at lower prices.
A Chinese model’s benchmark claims
In posts on X, DeepSeek said V4.1-Flash-1017 has “stronger benchmark performance, faster task completion, and shorter outputs than the previous version.” The company described it as an update to V4.1-Flash, which it released in September.
A technical report on Hugging Face, an online platform for AI models, data sets and apps, said the latest iteration made “substantial improvements across reasoning, coding, agentic tasks, and visual benchmarks.” DeepSeek called the release a minor update, with improvements covering uses from game creation to 3D modelling.
Nikkei Asia described the new version as outperforming OpenAI and Anthropic models by some measures. DeepSeek’s published results on Humanity’s Last Exam showed scores of 36.8 to 42.6 for the updated model, compared with 36.8 to 39.1 for V4.1-Flash. The exam measures performance across thousands of questions in math, sciences and humanities.
DeepSeek disclosed the figures on X. The base article offers no independent test results and no details on how the benchmark scores compare across all available models.
Price and competitive pressure
The update adds to DeepSeek’s challenge to Silicon Valley competitors Anthropic PBC and OpenAI, along with Chinese firms Moonshot Kimi and Z.ai. China’s open-weight models are competing with proprietary U.S. models described as top performers, though the latter carry far higher costs.
DeepSeek had offered its earlier V4.1-Flash system at a steep discount. After the company unveiled that model in early September, shares of MiniMax Group Inc, Z.ai and Alibaba Group Holding Ltd plummeted on the Hong Kong stock exchange. According to the report, Alibaba is driving its Qwen models.
Users can access the new model on DeepSeek’s website, app and developer platform. The company says the latest update builds on V4.1-Flash rather than introducing a wholly separate system.
Service strain remains
DeepSeek also experienced an outage that lasted about five hours and left its servers unreachable. Some experts blamed insufficient computing resources for the glitch.
Slowdowns and service disruptions had increased in recent months as developers sought DeepSeek’s aggressively priced models, putting pressure on its services. The source doesn’t identify any national policy change, migration impact or public opposition tied to the release. It also provides no information about a Huawei chip strategy, despite that phrase appearing in the topic title.
The available facts center on model performance claims, pricing and pressure on DeepSeek’s infrastructure from demand. The company is competing against Chinese and U.S. developers, with its published benchmark results and low-cost offering forming the basis of its latest challenge.
- deepseek
- v4.1-flash-1017
- artificial intelligence
- ai models
- openai
- anthropic
- china
- ai benchmarks