MiniMax-M3
428B params · 1M context · MoE
Online
MiniMax-M3
A native multimodal model with 1M token context, ~428B parameters (~23B activated), and MiniMax Sparse Attention — delivering 9× prefill & 15× decode speedups at 1M context. Supports text, images, and video.
🧠 Key capabilities
🗽 Analyze an image
⚡ Explain MSA
💻 Write code
Press Enter to send · Shift+Enter for new line