amdenterpriseai/aim-minimaxai-minimax-m3 repository overview

⁠MiniMax M3

MiniMax-M3 is a native multimodal mixture-of-experts model with 428 billion total parameters and 23 billion active parameters. It supports text, image, and video input, a 1 million-token context window, reasoning, tool use, and coding. MiniMax Sparse Attention reduces long-context compute, and the MXFP8 variant provides a smaller deployment footprint.