Chehre - Distributional Coverage (High Ambiguity, Random Sampling): leaderboard
Distributional expression recognition on Chehre (Azari et al., 2026): the model rates every candidate label of a video 25 times with random sampling (temperature 0.7), and coverage is the percentage of individual human rating vectors that at least one sampled vector falls near. Computed on the 300 videos on which perceivers disagreed most; higher is better.
Metric: Coverage (%; share of the human annotators' rating vectors that at least one of the model's 25 randomly sampled rating vectors falls within, using each human vector's distance to its 5th-nearest human neighbour as the radius; 300 highest-ambiguity videos, temperature 0.7, top-p 0.9, top-k 20). Source: arxiv.org. Saturation forecast: Around 2029. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen2.5-Omni-7B | 76 | #1797 |
| 2 | Gemini 2.5 Flash | 56 | #452 |
| 3 | GPT-5.6 Terra (Non-reasoning) | 40 | #60 (GPT-5.6 Terra) |
| 4 | InternVL3-8B | 27 | #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-distributional-coverage-high-ambiguity-random-sampling · How It Works · Data refreshed daily, snapshot 2026-11-12.