Pause-and-Think-B: leaderboard

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 May 2027. 11 models tracked.

Top models

#ModelScore
1GPT-5.264.24
2Qwen 3 VL 235B A22B Instruct58.89
3Gemini 2.5 Pro55.39
4GPT-4o50.33
5Qwen 3 VL 4B Instruct49

Interactive version: theaggregate.ai/benchmark?slug=pause-and-think-b · How It Works · Data refreshed daily, snapshot 2026-09-29.