Low-Cost DeepSeek AI Pressures Tech Rivals

DeepSeek released an updated low-cost AI model on Oct. 17, saying it improved performance as it competes with higher-cost proprietary tools from OpenAI and Anthropic. The Hangzhou-based startup offers V4.1-Flash-1017 through its website, app and developer platform. China’s open-weight models are competing with U.S. systems and domestic rivals.

The Cost Divide

DeepSeek called V4.1-Flash-1017 a minor update to V4.1-Flash, which it unveiled early in September. In posts on X, the company said the latest version had “stronger benchmark performance, faster task completion, and shorter outputs than the previous version.” Its technical report on Hugging Face said it made “substantial improvements across reasoning, coding, agentic tasks, and visual benchmarks.” DeepSeek said the model can be used for tasks ranging from game creation to 3D modelling.

The contest pits Chinese open-weight models against proprietary U.S. models, which the article describes as top performers that come with far higher costs. DeepSeek’s lower-cost approach challenges Silicon Valley companies Anthropic PBC and OpenAI, along with domestic competitors Moonshot Kimi and Z.ai. The article doesn’t provide prices for V4.1-Flash-1017 or rival systems.

DeepSeek reported results from Humanity’s Last Exam, a benchmark made up of thousands of questions in math, sciences and humanities. V4.1-Flash-1017 scored between 36.8 and 42.6, compared with 36.8 to 39.1 for V4.1-Flash. The startup said the newer model improved on the previous release. Nikkei Asia described it as outperforming OpenAI and Anthropic models by some measures.

Market Pressure and Service Strain

DeepSeek offered the earlier V4.1-Flash at a steep discount. After its release, shares of MiniMax Group Inc, Z.ai and Alibaba Group Holding Ltd fell sharply on the Hong Kong stock exchange. Alibaba is driving its Qwen models. The article links the market reaction to the discounted model’s arrival, but gives no share-price figures.

Model launches keep coming fast. Competition between open-weight and proprietary systems has intensified, and DeepSeek’s update extends its contest with Moonshot Kimi and Z.ai in China. The article reports no worker or union response to the launch. It also gives no figures for employment, wages or company profits.

Developers have sought DeepSeek’s aggressively priced models, and demand has put pressure on the company’s services, according to the article. Slowdowns and service disruptions increased in recent months. DeepSeek also experienced an outage that left its servers unreachable. Some experts attributed the glitch to insufficient computing resources.

The model is available through DeepSeek’s web service, app and developer platform. DeepSeek’s claims focus on benchmark performance, speed, shorter outputs and lower cost, while the reported disruptions show a constraint on delivering the tools at scale. The article gives no account of government action, labor conditions inside the company, or the costs of developing the model.