MiMo-v2.5-Pro-UltraSpeed: 1T model with 1000 tokens per second
MiMoAIspeedtrillion-parameter modelTileRTFP4DFlashinferenceultra-low-latencyheterogeneous execution system
Author: gainsurier
Date: 6/8/2026
Article Summary:
Xiaomi releases MiMo-V2.5-Pro-UltraSpeed, a 1T-parameter model that achieves 1000 tokens/s decode speed, breaking previous records in AI reasoning speed.