Event ·Multi-source

自主改进框架 Prime Agent 助 Opus 5 得分大涨，但现实价值存疑

Prime Agent 是一个能自主改进的递归 Agent 框架。替换外部 Harness 后，Opus 5 的 ARC-AGI-3 评分从 30.2% 升至 95.5%，超过人类专家基线 95.4%。但作者认为，这种面向评分的优化价值有限，在现实场景中不会比 Claude Code 和 Codex 更好。

Current read

这个框架把 Opus 5 的评分从 30% 刷到 95.5%，但作者觉得换汤不换药，实战未必比 Claude Code 强。

9 sources41° AI heatLast updated 2026/8/8 01:32:03

Evidence

9 sources

Primary source宝玉

2026/8/8 01:32:03View source

Related source 1小互

2026/8/6 13:30:27View source

Related source 2marktechpost

2026/8/6 09:00:08View source

Related source 3Claude

2026/8/7 03:08:54View source

Related source 4lmarena.ai

2026/8/7 18:20:47View source

Related source 5Decoder

2026/8/7 17:35:45View source

Related source 6Browser Use

2026/8/8 00:47:17View source

Related source 7量子位

2026/8/9 03:25:00View source

Related source 8Cursor

2026/8/6 15:40:11View 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