MiniMax Drops M3, an Open 428B Model With 1M-Token Context
MiniMax open-sources M3-MXFP8, a 428B-parameter multimodal MoE with a 1M-token context window, new sparse attention, and frontier coding scores — all runnable locally.
MiniMax-M3-MXFP8 is now live on Hugging Face: a quantized, locally-runnable version of MiniMax's new flagship model that packs 428 billion total parameters, a 1-million-token context window, native image and video understanding, and frontier-level coding performance into a single open-weight checkpoint. It is a significant moment for the open-weight ecosystem , and a direct challenge to the assumption that combining all three of those capabilities requires a closed, proprietary system.
MiniMax M3 is the first open-weight model to combine frontier-level coding performance, a 1-million-token context window, and native multimodal capabilities (image and video understanding) in a single architecture. The MXFP8 variant released on Hugging Face is a quantized version of the base model, designed to reduce memory overhead and make local deployment practical for teams with access to modern GPU hardware.