Pyyan / Open-Weight Models / MiMo-V2.6-Pro

X

MiMo-V2.6-Pro

A trillion parameters under the MIT licence.

currentXiaomi ↗Verified 2026-09-29

1.02T total

parameters, MIT

Specifications

Context1Msource · as of 2026-09-29
Input ($/Mtok)Weights onlysource · as of 2026-09-29
Output ($/Mtok)Weights onlysource · as of 2026-09-29
LicenceMITsource · as of 2026-09-29
Max outputNot publishedsource · as of 2026-09-29 · medium confidence
Parameters1.02T total, 42B activesource · as of 2026-09-29
Released21 Sep 2026source · as of 2026-09-29
Model IDXiaomiMiMo/MiMo-V2.6-Prosource · as of 2026-09-29 · medium confidence

What matters

  • 1.02T total parameters, 42B active per token, a 4.1% activation ratio
  • Text, image, video and audio in, with a 1M token context
  • Shipped with 7,000+ reinforcement learning environments and the framework that trained it

Best for multimodal work on your own cluster.