Xiaomi MiMo releases Pro, Omni models

Xiaomi is aiming at the part of the market that matters most now: models that can plan, use tools, and eventually perceive and act across modalities.

• –MiMo-V2-Pro is positioned for long-chain coding and OpenClaw-style workflows, which puts it in the same fight as other agent-first frontier models
• –MiMo-V2-Omni adds vision and audio, which matters if Xiaomi wants agents that operate in real-world, multimodal settings
• –The official homepage pushes both web demo and API access, so this looks like a developer-facing platform play, not just a benchmark splash
• –The real test will be reliability, latency, and integration quality, because “agentic” and “omni” are crowded claims now
• –If Xiaomi keeps shipping across text, audio, and multimodal, MiMo could become a broader AI stack instead of a one-off model drop