MiniMax M3

This model is coming soon to MyTokenGate. The details below are based on public vendor information; availability and pricing are governed by the model marketplace .

Overview

MiniMax M3 is MiniMax’s open-weights multimodal flagship, bringing coding, agentic reasoning, million-token context, and native multimodality together in one model. Built on a proprietary sparse-attention architecture, it natively supports million-token context windows for long-horizon agents, large codebases, and long-video understanding, and natively accepts text, image, and video input.

Key Features

• Open Multimodal Flagship: Coding, agentic reasoning, and multimodality in one
• Sparse Attention: Proprietary architecture for efficient long-context inference
• Native Multimodal: Text, image, and video input

Best Use Cases

• Long-Horizon Agents: Multi-step autonomous tasks and tool use
• Large Codebases: Project-level code understanding and generation
• Long-Video Understanding: Long-context multimodal analysis

Specifications

Capabilities and Limitations