MiniMax H3 open-source multimodal AI model gains immediate hardware acceleration

MiniMax made its H3 generation model publicly available, presenting a unified architecture for processing and producing content across text, images, audio, and video. The model creates native 2K video up to 15 seconds long and delivers commercial-grade results for diverse content workflows. External benchmarks placed H3 atop video editing rankings globally, while production costs remain competitive at 0.8 yuan per second. Hardware maker 摩尔线程 achieved same-day support, optimizing the model for its MTT S5000 accelerators and completing the entire integration within three hours.