Xiaomi MiMo-V2.6 Fixes Repeated Tool-Call Bug Caused by RL Reward Blind Spot

Xiaomi's MiMo team diagnosed and fixed a bug where MiMo-V2.6 models repeatedly called the same tools, burning context and stalling tasks; the root cause was a reinforcement learning reward blind spot, and the fix reportedly cost only about 4%.

2026-09-28 ~ 2026-09-28 · 1 related posts

• Xiaomi fixes MiMo-V2.6 tool-call repetition: an RL reward blind spot, patched at 4% of retrain cost — udmrzn · 2026-09-28