小米发布 MiMo-V2.6 论文：用强化学习走向自我改进
The paper for MiMo-V2.6 by Xiaom is out.
小米发布 MiMo-V2.6 论文，题为 MiMo-V2.6: Scaling Reinforcement Learning Towards Self-Improvement。论文中 AI 承担了大部分自身训练循环：智能体构建任务、审查测试、给答案打分并寻找作弊行为，人类只设定预算和规则。为应对通过/失败测试无法区分干净修复与取巧修复的问题，一个评分智能体比较同组内通过的补丁并把奖励移向更干净的方案；缺少它时智能体会偏向更长运行和吞掉异常等取巧做法。MiMo-V2.6-Pro 的 DeepSWE 分数在 260 万美元 RL 训练中从 58.4 升至 72.6，训练停止时仍在上升。
正文
当前显示原文，中文全文译文尚未就绪。
The paper for MiMo-V2.6 by Xiaom is out.
In MiMo-V2.6, AI runs much of its own training loop: agents build the tasks, audit the tests, grade the answers and hunt for cheats. Humans set the budget and the rules.
shows that agent models kept improving with more RL compute by scaling batch size, task and harness variety, and grading effort together.
Scaling RL for coding agents is hard because pass/fail tests can't tell a clean fix from a hacky fix. Agents also learn to game environments, for example by downloading the published fix.
A grader agent compared passing patches in each group and moved reward to the cleaner ones. Without it, agents drifted toward longer runs and workarounds like swallowed exceptions.
MiMo-V2.6-Pro’s DeepSWE score rose from 58.4 to 72.6 over $2.6M of RL and was still climbing when training stopped.
– arxiv. org/abs/2610.11959
Title: "MiMo-V2.6: Scaling Reinforcement Learning Towards Self-Improvement"