Chinese artificial intelligence startup DeepSeek has released a significant upgrade to its flagship model, DeepSeek-V3.1, highlighting a new feature optimized for “soon-to-be-released next-generation domestic chips.” This focus on compatibility with Chinese semiconductors signals a strategic positioning to align with Beijing’s push to develop a self-reliant technology ecosystem in the face of U.S. export restrictions on advanced processors.
The upgrade promises faster processing speeds and introduces a more efficient data format known as FP8 (8-bit floating point), which allows AI models to operate using less memory while running faster. The company, which shook the tech world this year by releasing models that compete with Western offerings like ChatGPT, did not specify which Chinese chip manufacturers or models would be supported by this new optimization.
Beyond hardware compatibility, the DeepSeek-V3.1 model features a new hybrid inference structure that allows it to operate in both standard and enhanced “deep thinking” modes. Users can toggle between these reasoning and non-reasoning modes via a button on the company’s official app and web platform, which have now been updated to the latest version.
This marks the company’s third model update this year, following releases in March and May. In a move that will affect developers, DeepSeek also announced it will adjust the pricing for using its API platform, which allows other applications to integrate its AI technology, starting September 6th.
