ESCUCHA - Audio Instruction Following: leaderboard
ESCUCHA (2026) is a Spanish speech-understanding benchmark of 1,000 expert-curated questions over in-the-wild recordings from seconds to 85 minutes long, with multiple accents and non-normative speech. Pass rate in percent on 100 items whose instruction is spoken in the audio and checked by a deterministic verifier; higher is better.
Metric: Pass rate (%; 100 open-ended items whose instruction (keywords, length, word exclusion, openings or closings, formatting, style) is spoken in the audio, checked by a deterministic verifier; in-the-wild Spanish audio from 6.7 s to 85 min per question with multiple accents and non-normative speech; one shared system prompt for every model). Source: arxiv.org. Saturation forecast: Around February 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemma 4 12B (IT) | 60 | #693 |
| 2 | Gemini 2.5 Flash | 54 | #453 |
| 3 | Qwen2.5-Omni-7B | 54 | #1789 |
| 4 | Voxtral-Mini-3B-2507 | 40 | #3673 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=escucha-audio-instruction-following · How It Works · Data refreshed daily, snapshot 2026-11-13.