AudioBench - TED-LIUM 3 Long Form WER: leaderboard
AudioBench TED-LIUM 3 Long Form word error rate slice measuring automatic speech-recognition transcription quality, where lower scores indicate fewer errors.
Metric: WER (%) (lower is better). Source: github.com. Saturation forecast: Estimated already saturated. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Whisper Large-v3 | 3.21 | #1271 |
| 2 | Phi-4 Multimodal Instruct | 3.82 | #906 |
| 3 | Qwen2-Audio-7B-Instruct | 8.74 | #880 |
| 4 | SALMONN-7B | 14.23 | #1293 |
| 5 | Qwen-Audio-Chat | 29.12 | #1244 |
| 6 | SeaLLMs-Audio-7B | 32.09 | #1018 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=audiobench-ted-lium-3-long-form-wer · How It Works · Data refreshed daily, snapshot 2026-10-20.