Chinese artificial intelligence startup DeepSeek has released an updated version of its flagship V3 model. According to Reuters, the new V3.1 introduces optimization for Chinese-made chips and delivers improved data processing speed.
The update supports the UE8M0 FP8 precision format, which allows AI models to run faster and more efficiently by reducing memory requirements. DeepSeek stated that V3.1 has been specifically optimized for the next generation of domestic processors.
A key feature of the release is its hybrid output structure, enabling users to switch between “reasoning mode” and “non-reasoning mode.” In the official DeepSeek app and web platform, a new “deep thinking” button now provides access to this functionality.
In recent months, the company has been rolling out frequent updates: in May, it launched an enhanced version of its R1 model, and earlier this year, it refreshed the original V3. The arrival of V3.1 once again underscores DeepSeek’s ambitions to compete with Western rivals such as OpenAI’s ChatGPT, while offering lower operating costs.
Starting September 6, the company will also adjust its API pricing, which allows developers to integrate DeepSeek into their own products.
