Chinese artificial intelligence startup DeepSeek released an upgraded version of its flagship V3 model on Thursday, introducing a new feature that optimizes performance for Chinese-made chips and delivers faster processing speeds.

The company said in a WeChat post that the updated DeepSeek-V3.1 model uses a UE8M0 FP8 precision format, which is tailored for “soon-to-be-released next-generation domestic chips.”

It did not specify which chip models or manufacturers the model would support. FP8, or 8-bit floating point, is a data format that enables AI models to function more efficiently by reducing memory usage and increasing speed compared to traditional formats.

The update follows two other recent model enhancements earlier this year, including an R1 update in May and a previous V3 improvement in March.

DeepSeek-V3.1 also includes a hybrid inference structure that allows the model to function in both reasoning and non-reasoning modes.

According to the company, users can switch between these modes using a “deep thinking” button available on its official app and web platform, which now operates on the new version.

In addition, DeepSeek announced changes to its API pricing structure, which allows third-party apps and web platforms to integrate the company’s AI model. These adjustments will take effect starting September 6.

DeepSeek gained global attention earlier this year by releasing AI models positioned as competitive alternatives to Western systems such as OpenAI’s ChatGPT, while offering lower operational costs.

The company’s increased focus on compatibility with Chinese-made chips comes amid growing efforts in Beijing to develop a domestic semiconductor ecosystem in response to ongoing export restrictions from Washington.

PHOTO: UNSPLASH

This article was created with AI assistance.