Ego-Exo4D Proficiency (Ego+Exo, ProfVLM Study): leaderboard
Metric: Top-1 accuracy (%) for four-tier demonstrator proficiency on the 466-video Ego-Exo4D validation split; Ego+Exo views; general VLMs zero-shot, ProfVLM variants trained as distinct specialist checkpoints; random 24.9%, majority-class 31.1%; higher is better. Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | ProfVLM-AGP | 48.2 | |
| 2 | ProfVLM-MLP | 42.9 | |
| 3 | SmolVLM2-256M | 32.2 | |
| 4 | SmolVLM-256M | 31.1 | |
| 5 | PandaGPT-13B | 25.8 |
Interactive version: theaggregate.ai/benchmark?slug=ego-exo4d-proficiency-ego-plusexo-profvlm-study · How It Works · Data refreshed daily, snapshot 2026-10-12.