MiMo-V2.5
Native omnimodal MoE model from Xiaomi supporting text, image, video, and audio understanding with agentic capabilities. 310B total / 15B active parameters, trained on ~48T tokens with hybrid sliding-window + global attention.
Parameters
310.8B
Context
1000000
Modality
text,image,video,audio->text
License
mit
Open source
Yes