Event ·Multi-source

ActiveVision新基准:GPT-5.5和Claude Fable 5主动观察得分不足11%

多模态大模型(MLLM)在主动观察任务上能力缺失。新基准ActiveVision包含17个任务(3个类别),要求模型反复视觉感知而非单次描述。GPT-5.5在最高推理档次上仅解决10.6%的项,11个任务得零分;Claude Fable 5仅3.5%,而三名人类被试平均96.1%。即使让模型编写并运行自己的视觉代码,大部分差距依然存在,因为代码在真实图像上不可靠,且捕捉自身失败需要模型缺乏的主动感知。结果表明当前MLLM缺乏稳健的主动视觉观察能力。

Current read

想看最强的AI在实时观察任务上有多弱吗?ActiveVision让GPT-5.5和Claude Fable 5原形毕露,人类碾压它们。

11 sources34° AI heatLast updated 2026/7/17 17:46:23

Evidence

11 sources
Related source 2lmarena.ai
View source
Related source 3Aravind Srinivas
View source
Related source 4shao__meng
View source
Related source 5Fireworks AI
View source
Related source 7Amjad Masad
View source
Related source 10Jerry Liu
View 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.