MiMo-V2.6 Bug Triage: Screenshot, Voice Note and Logs

Xiaomi shipped MiMo-V2.6 on September 22, and within a day it was the most talked-about open-weights release of the week. Artificial Analysis put MiMo-V2.6-Pro at 46 on its Intelligence Index, the top open-weights score at launch, ahead of Kimi K3 and Qwen3.8 Max. The weights are MIT-licensed on Hugging Face, and a Hacker News thread on the launch quickly turned into a wider argument about closed-model quotas and how hard it is to switch providers.

Most of the launch coverage stopped at benchmarks. The more useful detail for builders is that Pro and Flash are natively omnimodal: one chat request can carry an image, an audio clip, video and text, and the model reasons over all of them in the same pass, with tool calling on top. You don't need a separate speech-to-text step, a vision captioner, and then a text model to stitch the pieces together.