跳到正文
ARC Prize· @arcprize · X·· 1 小时前AI 评分33
AI 导读

在标准测试框架下以 max 推理运行时，55 个半私有游戏中有 28 个达到了 12 小时的运行时限，限制了测试能推进的程度。max 在 ARC-AGI-3 上得分 65.7%，而 xhigh 为 81.2%；若无时限限制，我们预计 max 的得分会更高。 另请注意，已发布的 Fable 5.2 在 medium 和 xhigh 推理强度下使用的调优与我们此前评估的检查点不同。low、high 和 max 未变。

正文 · 原文

At max reasoning with the standard harness, 28 of 55 semi-private games reached the 12-hour runtime limit, capping how far testing could progress. Max scored 65.7% on ARC-AGI-3 versus 81.2% at xhigh; without the limit, we expect max would have scored higher.

Also note that the released Fable 5.2 uses different tuning at medium and xhigh reasoning effort than the earlier checkpoint we evaluated. Low, high, and max are unchanged.

来源：ARC Prize · x.com

推理能力评测基准

事件后续 · 5

查看事件全部