MiniMax M3

About this model

MiniMax M3 is a 428B‑parameter multimodal language model with a sparse MoE architecture that supports text, image, and video inputs and generates text. It features a 1‑million token context window enabled by MiniMax Sparse Attention, delivering 9× faster prefill and 15× faster decode compared to its predecessor. The model achieves frontier‑level performance on long‑horizon agentic benchmarks and excels in coding and cowork tasks. It is released under an Other license.

Technical specifications

Capabilities

Input modalities

Output modalities

Reasoning

Hybrid Default on

Knowledge horizon