DeepSeek launches upgraded V3.1 AI model optimized for Chinese chips

Chinese AI startup DeepSeek released an upgraded version of its flagship V3 model on Thursday. The company stated that the update offers compatibility with Chinese-made chips and faster processing speeds.

The focus on local chip compatibility positions DeepSeek’s models to align with Beijing’s efforts to replace US technology in response to Washington’s export restrictions, supporting China’s growing semiconductor ecosystem.

Earlier this year, DeepSeek shook the tech world by launching AI models that rivaled Western systems like OpenAI’s ChatGPT while offering lower operating costs.

The V3 model upgrade follows two earlier core model updates: the R1 model in May and a previous V3 improvement in March.

In a WeChat post, DeepSeek said the DeepSeek-V3.1 model is optimized for the UE8M0 FP8 precision format, designed for “upcoming new-generation domestic chips.” The company did not disclose which chip models or manufacturers will be supported.

FP8, or 8-bit floating point, is a data format that allows AI models to run faster and use less memory compared to traditional methods, making them more efficient.

The company also said on Thursday that the DeepSeek-V3.1 model uses a hybrid inference architecture, enabling it to operate in both reasoning and non-reasoning modes. Users can switch between these modes via the “deep thinking” button on its official app and web platform, both of which now run on version V3.1.

The announcement added that starting September 6, the company will adjust the costs of using the model’s API, which allows developers of other apps and web products to integrate its AI models.