China’s DeepSeek drops upgraded V3.1 model optimized for Chinese-made chips

Chinese AI startup DeepSeek has unveiled its upgraded model, DeepSeek-V3.1, which the company says runs faster than its predecessor and is optimized for Chinese-made chips. The new version introduces a hybrid inference structure that lets users switch between reasoning and non-reasoning modes through a “deep thinking” button available on the company’s app and web platform. Early reviews have been highly positive, praising the model’s advanced reasoning and problem-solving abilities, especially in mathematics and coding, where it has shown the ability to break down complex tasks and even create simple games and applications from scratch. DeepSeek, which previously gained attention for offering competitive AI models at much lower operational costs than Western rivals such as ChatGPT, continues to emphasize affordability with this release. The V3.1 model also supports the FP8 precision format, designed for upcoming domestic chips, underscoring the company’s alignment with China’s growing semiconductor ecosystem, although no specific manufacturers were mentioned.

Credit: CGTN