DeepSeek has introduced an upgraded version of its V3 model, highlighting new features aimed at supporting future Chinese-made chips. The development, announced on Thursday, is being viewed as part of Beijing’s strategy to build technological self-sufficiency.
The improved system, called DeepSeek-V3.1, uses FP8 precision format, a technique that helps reduce memory use and speeds up AI processing. The company said the model was prepared to work with “next-generation domestic chips,” though it did not specify which brands or products would be supported.
DeepSeek has quickly become a prominent player in global AI after its models presented strong competition to Western systems such as OpenAI’s ChatGPT. Its products, known for being cheaper to operate, have made the startup a serious rival in the fast-expanding sector.
The latest upgrade comes after two recent updates earlier this year, reflecting the company’s accelerated development cycle. Beyond hardware compatibility, the V3.1 introduces a hybrid inference mechanism, enabling it to operate differently depending on user needs. A “deep thinking” button has been added to DeepSeek’s app and website, allowing users to toggle between reasoning-heavy and lighter tasks.
The firm also revealed that API pricing, which lets outside developers embed the model into digital platforms, will be revised from 6 September. The move is expected to attract more developers while broadening DeepSeek’s commercial reach.