Chinese artificial intelligence Startup Deepseek revealed an upgrade to its flagship V3 model on Thursday for which the company has announced that it has a function that can optimize chips made, together with higher processing speeds.
The focus on domestic chip compatibility can indicate that the AI models of Deepseek are positioned to collaborate with the emerging semiconductor ecosystem of China, while Beijing insists to replace the American technology in the light of Washington’s export restrictions.
Deepseek shook the technology world this year when the AI models released that compete with Western such as OpenAi’s Chatgpt and at the same time offer lower operational costs.
The upgrade to the V3 model of Deepseek follows two other recent updates for its core models -an R1 model update in May and an earlier V3 improvement in March.
For domestic chip support, Deepseek said his Deepseek-V3.1 model’s UE8M0 FP8 Precision format in a WeChat Post has been optimized for “soon within the generation of the next generation of domestic chips”.
The company did not determine which specific chip models or manufacturers would be supported.
FP8, or 8-bit floating point, is a data processing format with which AI models can work more efficiently, using less memory while they are performed faster than traditional methods.
The Deepseek-V3.1 has a hybrid inference structure with which the model can work both in reasoning and non-recurring modes, the company said in a WeChat post on Thursday.
Users can switch between these modes using a button “Deep Thinking” on the official app of the company and the web platform, both of which now perform the V3.1 version.
The company will also adjust the costs for the use of the API of the model, a platform with which developers from other apps and web products can integrate AI models, from 6 September, the statement showed.
