ActiveVision benchmark finds frontier vision models far behind humans at repeated visual reasoning
OfirPress · x · 2026-07-24
ActiveVision is a new benchmark for testing whether models can repeatedly observe, reason, and seek new visual evidence instead of making a single glance decision.
- Humans solve 96.1% of the tasks.
- The best frontier model reaches only 10.6%.
- Even strong models such as Fable 5 score just 3.5%.
The result suggests current vision models are still far from robust multi-step visual reasoning in interactive settings.
Related event: ActiveVision Benchmark Reveals Frontiers in Vision Models Lag Behind Humans(3 posts)→
More from Research
- On-Policy Self-Training paper: self-generated training data causes diversity collapse — idanbeck · 2026-10-27
- EMNLP 2026: sub-billion-parameter LMs show superhuman entity tracking, human-like fleeting memory — GretaTuckute · 2026-10-27
- 10-year study documents first 'unnatural gradient' phonotactic restriction in Tarma Quechua — begusgasper · 2026-10-27
- Clinical System One v0.2: top clinical AI drops from 97% to 65% on harder cases — MaziyarPanahi · 2026-10-27
- IdeaAnchor mines 14K papers to teach LLMs literature-grounded ideation — armancohan · 2026-10-26
- MIT's Ziming Liu: AutoResearch doesn't generalize — blind training tricks cause 'forking' disasters — ZimingLiu11 · 2026-10-26