Chehre - Recall@2: leaderboard
Top-two facial-expression recognition on Chehre (Azari et al., 2026): the percentage of the two highest-rated human labels for each of 2,111 videos that the two labels predicted by the model recover, ignoring order. Higher is better.
Metric: Recall@2 (%; share of the two highest-rated human labels recovered by the model's two predicted labels, ignoring order, averaged over the 2,111 videos). Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 60.7 | #452 |
| 2 | GPT-5.6 Terra (Non-reasoning) | 59 | #60 (GPT-5.6 Terra) |
| 3 | Qwen2.5-Omni-7B | 42.9 | #1797 |
| 4 | InternVL3-38B | 40.2 | #905 |
| 5 | InternVL3-8B | 28.1 | #1695 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=chehre-recall-2 · How It Works · Data refreshed daily, snapshot 2026-11-12.