PolySpeech-100: leaderboard
Spoken reading comprehension in 110 languages and dialects: the model hears a Belebele passage, question and four options (human recordings, dialect speech synthesised with CosyVoice 3, or neural TTS) and answers with a letter; PolySpeech-100 (Yang et al., KDD 2026). Accuracy in percent (chance 25%), macro-averaged over languages, with the task given as a text system prompt; higher is better.
Metric: Accuracy (%; spoken four-option reading comprehension from Belebele: the audio carries the passage, question and options; text instruction as system prompt; mean over all 110 languages and dialects). Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 85.3 | #141 |
| 2 | Qwen2.5-Omni-7B | 50.89 | #1797 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=polyspeech-100 · How It Works · Data refreshed daily, snapshot 2026-11-12.