Primary sourcerohanpaul_ai
View sourceEvent ·Multi-source
Google DeepMind 联合多所高校发布论文:GPT-6-Astra 解出 50 道未发表物理题中的 49 道
Google DeepMind 与多所高校的研究者收集了 50 道未发表的研究级理论物理题,每道题约需世界级专家一天完成。GPT-6-Astra 在 Codex 中配合网络搜索解出 49 道,单题成本不到 3 美元。GPT-6-Astra 与 Opus 5.5 在 Claude Code 中不使用搜索也各解出 46 道。是否联网对准确率影响很小,论文据此推测模型已记住大量物理文献。评分只看最终数值答案,区别于以往依赖教科书或竞赛题的基准。
Current read
GPT-6-Astra 在新基准里解出 49/50 道研究级物理题,每道不到 3 美元,还能对比联网和不联网的差距。
10 sources51° AI heatLast updated 2026/10/25 10:42:05
Evidence
10 sourcesRelated source 1Teortaxes
View sourceRelated source 2kimmonismus
View sourceRelated source 3Yuchen Jin
View sourceRelated source 4Simon Willison’s Weblog
View sourceRelated source 5IT之家
View sourceRelated source 6TestingCatalog
View sourceRelated source 7Bindu Reddy
View sourceRelated source 8SiliconFlowAI
View sourceRelated source 9Decoder
View sourceConflict 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.