DeepSek Releases V3.1 Model With Local Chip Support

DeepSek, releasing an update for their flagship model, V3 (photo: x @deepsseek ·)

JAKARTA An artificial intelligence (AI) company from China, DeepSee, released an update for their flagship model, V3, on Thursday, August 21. This latest model is claimed to have features optimized to support domestically made chips and offer higher processing speeds.

The focus on compatibility with domestic chips is seen as a signal that the DeepSek AI model is being directed to support China's growing semiconductor ecosystem, as Beijing's efforts to replace US technology amid export restrictions imposed by Washington.

DeepSeek is in the spotlight of the world of technology this year after launching an AI model that is able to compete with Western products such as OpenAI's ChatGPT, but with lower operational costs.

The update on the V3 model follows two other improvements to the company's core model, namely the R1 model update last May and an early V3 upgrade in March.

In an upload on WeChat, DeepSek said the DeepSek-V3.1 model uses the UE8M0 FP8 precision format optimized for the next generation of domestic chips to be released soon. However, the company did not specify any chip models or manufacturers to be supported.

FP8 or 8-bit floating point is a data processing format that allows AI models to operate more efficiently, use less memory, as well as run faster than traditional methods.

DeepSek-V3.1 is also equipped with a hybrid inference structure that allows models to work in reasoning and non-scattering modes. Users can switch between the two modes via the deep thinking' button on the company's official application and website platform, which now uses the V3.1 version.

The company will also adjust the cost of using this model's API, which is a platform for app developers and other web products to integrate DeepSeeek AI models, starting September 6.