SEA-SpeechBench - Timestamped Content Query (30-60 s): leaderboard
Metric: WER or CER (%; transcribe only the speech inside a queried time window of a long recording of 30-60 seconds, CER for languages without word boundaries; English instruction prompt; blank where the model cannot take audio of that length; can exceed 100; lower is better). Source: arxiv.org. Saturation forecast: Around October 2026. 13 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen3 Omni 30B A3B Instruct | 1.22 |
| 2 | Gemini 2.5 Flash | 2.64 |
| 3 | Qwen2.5-Omni-7B | 6.58 |
| 4 | GPT-4o Audio | 6.82 |
| 5 | Gemma 3n E4B (IT) | 7.38 |
| 6 | gemma-3n-E2B-it | 7.49 |
| 7 | Voxtral-Mini-3B-2507 | 7.74 |
| 8 | Phi-4 Multimodal Instruct | 14.9 |
Interactive version: theaggregate.ai/benchmark?slug=sea-speechbench-timestamped-content-query-30-60-s · How It Works · Data refreshed daily, snapshot 2026-09-26.