ESDR-Bench: leaderboard
Metric: Pairwise preference accuracy (%) over all benchmark pairs (micro average of the modality and colloquialness subsets); the evaluator is given both full multi-turn spoken dialogue episodes of a pair (zero-shot audio judges by prompt, dedicated evaluators by their own scoring) and must prefer the labeled one; pairs come from ESDR-Bench, a source- and metadata-stratified sample of the held-out split; chance 50; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 76.42 | #145 |
| 2 | Gemini 2.5 Flash | 61.81 | #237 |
| 3 | Qwen2.5-Omni-7B | 51.47 | #601 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=esdr-bench · How It Works · Data refreshed daily, snapshot 2026-10-11.