Primary sourceIT之家
View sourceEvent ·Multi-source
Epoch AI 发布家具组装基准 FAB,OpenAI GPT-6 Astra 准确率达 80%
Epoch AI 于 9 月 23 日发布家具组装基准 FAB,用 60 张宜家家具错误组装照片测试多模态模型的视觉与空间推理能力。OpenAI GPT-6 Astra 以 80% 准确率位居第一,Anthropic Claude Fable 5.1 和 Claude Opus 5 分别达到 70% 和 61%。相比之下,2025 年 11 月的领先模型 Claude Opus 4.5 准确率仅 28%,约 10 个月内提升显著。GPT-6 Astra 分析单张照片中位耗时约 3 分钟,比此前领先模型快约 2 至 10 倍。研究还指出,中国模型中表现最好的开源权重模型 Kimi K3 在该基准上落后国际前沿约 7 个月。
Current read
Epoch AI 搞了个新基准,让 AI 看照片对照说明书找宜家家具装错的地方,GPT-6 Astra 拿了 80 分,还顺带测出了各家模型的不同毛病。
11 sources73° AI heatLast updated 2026/9/26 10:18:15
Evidence
11 sourcesRelated source 1Guillermo Rauch
View sourceRelated source 2Artificial Analysis
View sourceRelated source 3Blognone
View sourceRelated source 4Decoder
View sourceRelated source 5lmarena.ai
View sourceRelated source 6elvis
View sourceRelated source 7kimmonismus
View sourceRelated source 8Riley Goodside
View sourceRelated source 9Jerry Liu
View sourceRelated source 10shao__meng
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.