推文

@yueke · 2026-10-18 00:07

This is how benchmarks should be done. Full methodology in the open, and anyone can rerun the numbers instead of arguing about a chart. More teams building agent evals should follow this playbook.

曝光 13 · 评论 0 · 点赞 0 · 书签 0 · 曝光/时 7.994879725916334