MiniMax M2
MiniMax · 2025-10-27 · 228.7B parameters
MiniMax M2 is MiniMax's approximately 229-billion-parameter MoE language model released in October 2025, built for agentic coding and long-horizon task workflows. It activates about 10 billion parameters per token, delivering the knowledge capacity of a large MoE model at lower inference cost, and its weights are documented in the MiniMax-M2 GitHub repository.