Event ·Official first-hand

Trace2Eval:从真实交互轨迹构建编程智能体评测基准

arXiv 论文 Trace2Eval 提出从真实用户交互轨迹动态构建编程智能体评测基准的流水线,替代依赖静态公共任务的现有 benchmark。该方法用多智能体重建系统恢复原始用户查询、还原工作区文件并重建等效执行环境,再通过标注者对齐迭代细化人工评分标准,并用源码与运行时证据校准 judge 智能体。基于 Trace2Eval 反馈,19 个 model-harness 配置平均得分提升 10.05 个百分点,其中 DeepSeek-V4-Pro 配合 Claude Code 最高提升 19.28 个百分点。

Current read

一组研究者把真实用户和编程智能体的交互记录变成可复用的评测集,19 个配置平均涨了 10 个点,评测和改进方法都能参考。

2 sources38° AI heatLast updated 2026/11/4 13:36:50

Evidence

2 sources
Primary sourcearXiv: DeepSeek
View source
Related source 1Thomas Wolf
View 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.