Pause-and-Think-B: leaderboard
Pause-and-Think-B is a 300-clip egocentric and exocentric video benchmark for assistive action suggestion (Advanced Micro Devices, 2026): given a short clip and a user query, a model gives a free-form contextual answer or next-step plan, scored 0 or 1 by a GPT-5.1 judge that checks visual grounding and action feasibility over three runs. This board reports Binary validity accuracy (%): the GPT-5.1 judge marks each free-form video-grounded answer or next-step plan correct or not over three runs, averaged over the 100 scene and 200 goal items; higher is better.
Metric: Binary validity accuracy (%): the GPT-5.1 judge marks each free-form video-grounded answer or next-step plan correct or not over three runs, averaged over the 100 scene and 200 goal items; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 | 64.24 | #176 |
| 2 | Qwen 3 VL 235B A22B Instruct | 58.89 | #522 |
| 3 | Gemini 2.5 Pro | 55.39 | #266 |
| 4 | GPT-4o | 50.33 | #741 |
| 5 | Qwen 3 VL 4B Instruct | 49 | #1279 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=pause-and-think-b · How It Works · Data refreshed daily, snapshot 2026-11-13.