A major player in the artificial intelligence world, DeepSeek, has just released an updated AI model. This Chinese developer states their new DeepSeek V3 upgrade specifically boosts the performance of chips made in China. It also promises faster processing speeds.
This new version is called DeepSeek-V3.1. It uses a special processing style known as UE8M0 FP8. This format is fine-tuned for the “next-generation domestic chips” that will be available very soon. DeepSeek hasn’t shared any specific names or details about these upcoming chips.
DeepSeek also mentioned that DeepSeek-V3.1 has a unique real-world operating structure for its “inference” capabilities. This allows the AI to work in both reasoning and non-reasoning modes. Users can easily switch between these modes with a “deep thinking” button. This feature will be available on both the company’s app and its web platform.
The FP8 format is a data processing method that makes AI models run much more efficiently. It helps them use less computer memory while still working faster. This is key for pushing AI performance forward.
This isn’t DeepSeek’s only recent update. The company has been busy. They released their R1 model in May. Before that, they put out an earlier update for the V3 model in March of this year.