Pro2Bench: leaderboard
Pro2Bench (Plan, Watch, Recover, 2026) tests proactive procedural assistants on egocentric and instructional video: at each decision point the model either stays silent or interrupts with guidance, evaluated zero-shot over six sets including the new EgoProactive recordings. The board is the Proactive Quality Score in percent, crediting correct silence and well-timed guidance weighted by a GPT-5.2 judge quality rating; higher is better.
Metric: Proactive Quality Score (PQS, 0-1 scaled to %): per decision point, 1 for correct silence, the GPT-5.2-judged guidance quality (rescaled 0-1) for a correctly timed interrupt, 0 for spurious or missed interrupts; zero-shot monolithic assistant (8 uniformly sampled frames per clip, greedy decoding for the interrupt or silent decision), mean over the six evaluation sets (EgoExo4D, EPIC-KITCHENS, HowTo100M, Ego4D, HoloAssist, EgoProactive); higher is better. Source: arxiv.org. Saturation forecast: Around 2032. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.6 27B | 37 | #341 |
| 2 | Gemini 3.1 Pro (Preview) | 35 | #94 |
| 3 | Llama 4 Maverick | 35 | #1099 |
| 4 | Claude Opus 4.6 | 32 | #102 |
| 5 | Qwen 3 VL 235B A22B | 30 | #479 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=pro2bench · How It Works · Data refreshed daily, snapshot 2026-11-13.