TMTPOST — MiniMax open-sourced its next-generation multimodal generative model MiniMax H3 on August 3, and Muxi Co. immediately confirmed Day-0 adaptation on its Xiyun C-series GPUs. MiniMax H3 is a general-purpose multimodal generation model that unifies understanding of text, images, video and audio contexts and produces native stereo audio-video clips of up to 15 seconds at 2K resolution. Independent rankings place it first in video-editing benchmarks, while its pricing at 2K resolution is less than one-third that of mainstream peers. Muxi’s Xiyun C-series GPUs, powered by the company’s self-developed MXMACA full-stack software, completed deep adaptation on the day of release, enabling efficient deployment and inference without the usual multi-week porting cycle. The same software stack has already delivered Day-0 support for more than two dozen leading Chinese models across language, multimodal, OCR and generation categories. Muxi, listed on the Shanghai STAR Market, designs domestic general-purpose GPUs that cover AI training-inference, scientific computing and graphics. The rapid H3 enablement continues a pattern of close collaboration with MiniMax and underscores the growing maturity of China’s GPU-plus-model co-optimization loop, allowing developers to run the newly open-sourced multimodal model on indigenous hardware from the first day of availability.