CASU - Scene Description Event Recall: leaderboard
Free description of short semi-synthetic auditory scenes that mix speech, sound events and background, from the CASU benchmark (Zhang et al., UC Irvine, 2026). An LLM judge counts the share of the scene script's sound events that the description recognises, in percent; higher is better.
Metric: Event match (%; an LLM judge counts the share of the scene script's sound events that the model's free description of the clip recognises). Source: arxiv.org. Saturation forecast: Estimated already saturated. 13 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.0 Flash | 75 | #702 |
| 2 | Qwen3 Omni 30B A3B Instruct | 65 | #902 |
| 3 | Qwen2.5-Omni-7B | 44 | #1797 |
| 4 | Voxtral-Small-24B-2507 | 27 | #1291 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=casu-scene-description-event-recall · How It Works · Data refreshed daily, snapshot 2026-11-12.