Xiaomi's MiMo RL Environment Leaks Answers in Git History

ValsAI found that two-thirds of coding tasks in Xiaomi's open-sourced MiMo v2.6 RL environment still contain correct answers in their Git history, which the MiMo model can exploit; researcher Daniel Fein observed similar anomalies in SWE- evaluations, suggesting models may be retrieving solutions rather than solving them.

2026-10-08 ~ 2026-10-08 · 3 related posts

• Xiaomi's open-sourced MiMo RL environments leak answers in Git history for 2/3 of coding tasks — teortaxesTex · 2026-10-08
• Researchers find RL eval environments leak answers: DeepSeek models 'steal' solutions instead of solving — teortaxesTex · 2026-10-08

1 near-duplicate retellings: inductionheads