Event ·Multi-source

GPT-6 Astra 在 TheoPhysEval 物理评测中拿下 49/50

TheoPhysEval 的作者发现，GPT-6 Astra 在该物理基准上的部分答案被判错，实际是参考解答本身有误，复核后确认 Astra 是对的。最终 Astra 得分 49/50（98%）。评测过程中使用 Codex 配合工具调用和联网检索完成。

Current read

物理评测的标准答案自己都错了，GPT-6 Astra 反而是对的，最后拿 98%，这条挺有意思的。

11 sources64° AI heatLast updated 2026/10/15 04:58:00

Evidence

11 sources

Primary sourcekimmonismus

2026/10/15 04:58:00View source

Related source 1Decoder

2026/10/15 15:24:15View source

Related source 2Artificial Analysis

2026/10/13 18:00:33View source

Related source 3Epoch AI

2026/10/15 11:26:45View source

Related source 4DeepLearning.AI

2026/10/13 16:00:00View source

Related source 5Bindu Reddy

2026/10/13 16:01:39View source

Related source 6Ethan Mollick

2026/10/13 16:57:33View source

Related source 7TestingCatalog

2026/10/13 18:16:12View source

Related source 8IT之家

2026/10/13 23:44:48View source

Related source 9OpenAI Blog

2026/10/14 07:00:00View source

Related source 10berryxia

2026/10/14 08:28:24View 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