PIVOTS-Bench - Visual Cue Causal Analysis: leaderboard
From a key frame of a Social-IQ 2.0 clip, choose the two annotated visual cues (such as facial expression, gesture or posture) out of five that support a given interpersonal-relationship score; PIVOTS-Bench (Zhang et al., 2026). Half credit per correct cue, in percent (chance 40%); higher is better.
Metric: Score (%; choose the two annotated visual cues out of five that support a relationship score in a key frame; 0.5 credit per correct cue; chance 40%). Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 60.25 | #163 |
| 2 | Gemini 2.5 Pro | 48 | #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-visual-cue-causal-analysis · How It Works · Data refreshed daily, snapshot 2026-11-12.