Qwen-Audio-3.1-Realtime Report - LibriSpeech other: leaderboard
Qwen-Audio-3.1-Realtime report Table 4: native author evaluations. MMAU uses accuracy percent; LibriSpeech clean/other use WER percent; FLEURS is macro-average WER percent across supported languages (lower is better). LongBench v 2 and MiniLongBench use accuracy percent under source long-context evaluation. Model controls retain explicit source-scoped unspecified API effort and Qwen 3.0 Plus/Flash variant. Metric: Word error rate (%)
Metric: Word error rate (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen-Audio-3.1-Realtime | 2.31 | |
| 2 | Qwen-Audio-3.1-Realtime Report Qwen-Audio-3.0-Realtime (Plus or Flash unspecified) | 2.34 | |
| 3 | Qwen-Audio-3.1-Realtime Report Gemini 3.7 Flash (effort unspecified) | 6.46 |
Interactive version: theaggregate.ai/benchmark?slug=qwen-audio-3-1-realtime-report-librispeech-other · How It Works · Data refreshed daily, snapshot 2026-11-05.