SEA-SpeechBench - Timestamped Content Query (120-180 s): leaderboard

Metric: WER or CER (%; transcribe only the speech inside a queried time window of a long recording of 120-180 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. 7 models tracked.

Top models

#ModelScore
1Qwen3 Omni 30B A3B Instruct1.86
2Gemini 2.5 Flash4.09
3GPT-4o Audio8.59
4Voxtral-Mini-3B-250722.13

Interactive version: theaggregate.ai/benchmark?slug=sea-speechbench-timestamped-content-query-120-180-s · How It Works · Data refreshed daily, snapshot 2026-09-26.