Chinese AI startup DeepSeek has released an updated version of its flagship V3 model. The company says it is optimized to work with domestically produced chips and also provides higher data processing speeds.
Chinese AI startup DeepSeek has released an updated version of its flagship V3 model. The company says it is optimized to work with domestically produced chips and also provides higher data processing speeds.
The focus on Chinese chips is a signal that DeepSeek is preparing its AI models to work with the domestic semiconductor ecosystem, part of a broader strategy by Beijing to try to replace American technology through export restrictions, Reuters reported .
The update to DeepSeek's V3 model follows two other recent updates to its core models — the R1 model update in May and the previous V3 enhancement in March .
In a WeChat post, DeepSeek said its DeepSeek-V3.1 model is optimized to support domestic chips. It notes that the UE8M0 FP8 precision format is specifically designed for “new-generation domestic chips that will soon be released to the market.” It did not name specific chip models or their manufacturers.
Regarding domestic chip support, DeepSeek said in a WeChat post that the UE8M0 FP8 precision output format of the DeepSeek-V3.1 model is optimized for “next-generation domestic chips that will be released soon.” The company did not specify which specific chip models or manufacturers will be supported.
FP8, or 8-bit floating-point format, is a data processing format that allows AI models to run more efficiently, using less memory and running faster than using traditional methods.
DeepSeek-V3.1 has a hybrid inference structure that allows the model to operate in both reasoning and non-reasoning modes, the company said. Users can switch between these modes using the “deep thinking” button on the company’s official app and web platform, both of which are now live in V3.1.
The statement also states that starting September 6, the company will change the cost of using the model's API.
