Event ·Multi-source

ReTerm：自我修复终端智能体训练方法

ReTerm是基于Qwen3.5-9B构建的终端智能体，在Terminal-Bench 2.0基准测试上达到34.8%的Pass@1分数，比基础模型提升14.6个百分点。该模型通过ReTerm-10K数据集学习从失败中恢复，采用ReTerm-GRPO在线强化学习算法。研究显示，ReTerm在各类错误恢复率上均优于标准GRPO训练方法。

Current read

ReTerm能从自身错误中学习，终端任务恢复能力提升显著

3 sources39° AI heatLast updated 2026/10/13 14:44:24

Evidence

3 sources

Primary sourcearXiv cs.LG

2026/10/13 14:44:24View source

Related source 1DavidSacks

2026/10/14 15:45:04View source

Related source 2宝玉

2026/10/13 03:05:48View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story