Xiaomi streams two MiMo 2.6 RL runs

Xiaomi has opened a live dashboard for two MiMo-V2.6 reinforcement-learning runs—Pro and Flash—showing reward curves and evaluation telemetry as training proceeds, according to the public run dashboard. A Hacker News discussion of the release records Pro at 63.7% on DeepSWE 1.1 by step 10, but that is an in-run checkpoint, not an independent final benchmark. Builders should file the transparency move as useful process evidence—not model proof—and compare released checkpoints against fixed evaluations, the same separation between speed demos and capacity that Nvidia’s local-router analysis makes explicit.