PIVOTS-Bench - Key Frame Identification: leaderboard
Visual grounding of interpersonal-relationship judgements in Social-IQ 2.0 video clips: given a relationship dimension and its score, pick from five frames the one the annotators marked as best showing it (the distractors are frames 5 or 10 seconds away); PIVOTS-Bench (Zhang et al., 2026). Accuracy in percent (chance 20%), higher is better.
Metric: Accuracy (%; pick, from five frames of a Social-IQ 2.0 clip, the key frame the annotators marked as best showing a given relationship-dimension score; distractors 5 or 10 seconds away; chance 20%). Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 62.15 | #163 |
| 2 | Gemini 2.5 Pro | 46.63 | #262 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=pivots-bench-key-frame-identification · How It Works · Data refreshed daily, snapshot 2026-11-12.