DeepSeek has unveiled an updated AI model. It claims that it is faster than the previous one.

On Saturday, DeepSeek unveiled an updated version of its AI model, V4.1-Flash-1017. According to the company, it runs faster, provides shorter responses, and delivers better benchmark results than the previous version. The release intensifies the Chinese startup’s competition with U.S. giants Anthropic and OpenAI, as well as with local AI labs Moonshot Kimi and Z.ai, Bloomberg reports.

DeepSeek introduced the model in a series of posts on X. According to the developers, the updated version delivers “stronger benchmark results, faster task completion, and shorter responses compared to the previous version.” The service is already available via the web interface, the DeepSeek mobile app, and the developer platform.

In a technical report on the Hugging Face platform, the startup reported “significant improvements in reasoning, programming, agent-based tasks, and visual benchmarks.” The new version expands the model’s capabilities across a wide range of tasks—from game development to 3D modeling. The company itself describes it as a minor update to the previous V4.1-Flash version.

DeepSeek also released the results of the Humanity's Last Exam (HLE), designed to assess AI capabilities across thousands of questions in mathematics, the natural sciences, and the humanities. The updated model scored between 36.8 and 42.6 points, compared to 36.8–39.1 for the previous version.

The release came amid technical issues at DeepSeek. Earlier this week, the startup’s servers were down for about five hours. Some experts attributed the outage to a lack of computing power, Bloomberg notes. In recent months, service disruptions at the company have become more frequent amid growing demand from developers for its extremely low-cost models.

The V4.1-Flash model, which preceded the new version, was released in September at such a significant discount that the stock prices of DeepSeek’s competitors—MiniMax Group, Z.ai, and even the giant Alibaba—plummeted on the Hong Kong Stock Exchange. DeepSeek had long been known as one of the main “disruptors” in the AI market. In January 2025, the release of the R1 model—developed, according to the company, for less than $6 million—triggered a $1 trillion sell-off of U.S. and European tech stocks in a single day, including a 15% drop in Nvidia’s stock price.

At the same time, in August, DeepSeek sharply raised prices for its flagship V4 models. Bloomberg attributed this to the company’s preparations for an IPO and its efforts to boost profitability. Around that time, the company was also in talks regarding a funding round with a valuation of up to $71 billion.

This article was AI-translated and verified by a human editor