Skip to main content
MiniMax brings three agent-focused Mixture-of-Experts models to DeepMask. MiniMax M3 is the newest, with image input, hybrid reasoning, and a 1M-token context window. MiniMax M2.5 adds built-in reasoning for cost-efficient coding and tool use. MiniMax M2.1 is tuned for mobile app development and 3D visualization.

About

MiniMax M3 is the newest MiniMax model in DeepMask and the first in the family to read images. It is a 428B parameter Mixture-of-Experts model with hybrid reasoning and a 1M-token context window, which suits long-running agents, large codebases, and multimodal document work.

Specifications