GlobeAudio: leaderboard
GlobeAudio (2026) evaluates audio-language models on multiple-choice questions written by native speakers about naturally occurring audio in English, Russian, Chinese, Thai, Singlish and Bengali, requiring auditory reasoning and cultural interpretation. The score is the macro-average accuracy in percent over the six languages, with the audio fed to the model directly; higher is better.
Metric: Macro-average over six languages of accuracy (%) on GlobeAudio's naturalistic-audio multiple-choice questions written by native speakers; audio fed to the model directly, source-language prompts, predicted option label against the key. Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 82.68 | #94 |
| 2 | Gemini 2.0 Flash | 66.72 | #704 |
| 3 | MiMo-V2.5 | 61.08 | #354 |
| 4 | Voxtral-Small-24B-2507 | 57.94 | #1291 |
| 5 | Gemma 3n E4B (IT) | 56.53 | #2210 |
| 6 | MERaLiON-2-10B | 55.64 | #1640 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=globeaudio · How It Works · Data refreshed daily, snapshot 2026-11-13.