Chinese AI startup DeepSeek has announced that its new model V3.1 utilizes an "UE8M0 FP8" precision format, optimized for upcoming domestic AI chips. This aligns with China's push for self-sufficiency amid U.S. export restrictions. DeepSeek's advancements, such as faster response times and a hybrid reasoning architecture, come as Beijing encourages local developers to use domestic alternatives to Nvidia's GPUs. Starting September 6, usage fees for DeepSeek's API will be adjusted.
Chinese AI startup DeepSeek has announced that its new model V3.1 utilizes an "UE8M0 FP8" precision format, optimized for upcoming domestic AI chips. This aligns with China's push for self-sufficiency amid U.S. export restrictions. DeepSeek's advancements, such as faster response times and a hybrid reasoning architecture, come as Beijing encourages local developers to use domestic alternatives to Nvidia's GPUs. Starting September 6, usage fees for DeepSeek's API will be adjusted.