Chinese AI startup DeepSeek has released V3.1, an update to its flagship V3 model, positioning it as a rival to OpenAI’s GPT-5, marking a significant development in the global AI landscape.
DeepSeek’s V3.1 was quietly launched via a message on WeChat and the Hugging Face platform, underscoring China’s ambition to develop and control advanced AI systems independently of foreign technology.
DeepSeek’s models, including the R1 reasoning model, have seen widespread adoption in China and increasing interest globally, even among some American firms. However, concerns have been raised regarding the models’ outputs aligning with Chinese Communist Party narratives.
China’s AI sector is competitive, with models from companies such as Alibaba, Moonshot AI, and Baidu. DeepSeek’s V3.1 release, following OpenAI’s GPT-5, highlights China’s determination to compete with leading U.S. labs. OpenAI CEO Sam Altman acknowledged that competition from Chinese open-source models, including DeepSeek, influenced OpenAI’s decision to release its own open-weight models.
Altman stated, “It was clear that if we didn’t do it, the world was gonna be mostly built on Chinese open-source models. That was a factor in our decision, for sure. Wasn’t the only one, but that loomed large.” The U.S. government has granted licenses to Nvidia and AMD to export China-specific AI chips, with conditions that 15% of revenue from these sales goes to Washington.
In response, Beijing has restricted purchases of Nvidia chips following statements from Commerce Secretary Howard Lutnick indicating restrictions on sales of top-tier chips to China. Lutnick stated on July 15, “We don’t sell them our best stuff, not our second-best stuff, not even our third-best.” DeepSeek’s optimization for Chinese-made chips is a strategy to mitigate the impact of U.S. export controls.
The company mentioned in its WeChat post that the new model format is designed for “soon-to-be-released next-generation domestic chips.” Altman has expressed concerns about underestimating China’s progress in AI and suggested that export controls may not be a sufficient solution, remarking, “I’m worried about China.”
The V3.1 model is notable for its technical specifications, including 685 billion parameters. Its “mixture-of-experts” design lowers computing costs for developers by activating only a fraction of the model for each query. Unlike previous DeepSeek models, V3.1 combines both fast answers and reasoning in one system.
Tech analyst Ben Dickson, founder of the TechTalks blog, noted that V3.1’s hybrid architecture is “the biggest feature by far,” marking a significant advancement in AI technology.
