Chinese AI startup DeepSeek launches enhanced V3 AI model with China-made chip support and faster speeds

Chinese artificial intelligence startup DeepSeek has rolled out an upgraded version of its flagship model, underscoring the country’s broader effort to cut reliance on U.S. technology. The new release, called DeepSeek-V3.1, adds support for China-made chips and promises faster processing speeds.

While DeepSeek’s upgrade, shared in a WeChat post that isn’t publicly accessible, Reuters confirmed on Thursday that the company had released “an upgrade to its flagship V3 model that the company says has a feature that can optimize it for Chinese-made chips, along with faster processing speeds.”

Reuters added that “the focus on domestic chip compatibility may signal that DeepSeek’s AI models are being positioned to work with China’s emerging semiconductor ecosystem as Beijing pushes to replace U.S. technology in the face of Washington’s export restrictions.”

DeepSeek Unveils Upgraded V3 Model With Domestic Chinese Chip Support and Improved Performance

The release comes as Beijing ramps up efforts to strengthen its semiconductor base under tightening U.S. export controls. By optimizing its model for domestic hardware, DeepSeek is signaling that it wants to be at the center of that push.

The startup has already turned heads this year with AI models that rival Western systems like OpenAI’s ChatGPT while keeping operating costs lower. The V3.1 upgrade follows two earlier updates: an R1 refresh in May and a V3 enhancement in March.

In its announcement, DeepSeek said the model’s UE8M0 FP8 precision format is tuned for “soon-to-be-released next-generation domestic chips,” though it didn’t specify which manufacturers or models would be supported. FP8, or 8-bit floating point, is a data format that makes models more efficient by using less memory while processing at higher speeds.

The company also introduced a hybrid inference setup that lets the model switch between reasoning and non-reasoning modes. Users can toggle between them using a “deep thinking” button now available in DeepSeek’s app and web platform, both of which run on the updated model.

Alongside the technical upgrade, the startup said it will adjust pricing for developers using its API — which allows other apps and platforms to integrate its technology — starting September 6.

DeepSeek first made headlines earlier this year after launching an AI model built on older Nvidia chips at a fraction of the cost competitors were spending. The startup also released a global chatbot app that has already drawn millions of downloads, along with increased scrutiny from regulators. In late January, DeepSeek surpassed ChatGPT on the App Store as the Chinese AI startup sent shockwaves through tech stocks.

The buzz around DeepSeek intensified in February after reports that its V3 model had outperformed Meta’s Llama 3.1, OpenAI’s GPT-4o, and Alibaba’s Qwen 2.5 on third-party benchmarks — all while undercutting rivals on price.