Mimo livestreams full RL training of V2.6 Pro (1T, 42B active) with internal metrics

mishig25 · x · 2026-09-17

A viral repost highlights that Mimo is publicly livestreaming the RL training run of its V2.6 models: Mimo V2.6 Pro (1T total, 42B active) and Flash (309B, 15B active).

The stream exposes an unusually detailed metrics dashboard, including per-batch data/harness composition and a large set of internal training metrics that are normally kept private. Observers called it a rare full public view into a frontier-scale RL training run.

Related event: Xiaomi live-streams MiMo-V2.6 RL training, burning over $1.08M in 36 hours(14 posts)→

Original post →

More from Models

Models channel →