SEA-SpeechBench - Temporal Localization (60-120 s): leaderboard

Metric: Span-overlap F1 (%; predict the start and end time at which a queried phrase is spoken in a recording of 60-120 seconds; English instruction prompt; blank where the model cannot take audio of that length). Source: arxiv.org. Saturation forecast: Around 2031. 10 models tracked.

Top models

#ModelScore
1Qwen2.5-Omni-7B11.32
2GPT-4o Audio8.53
3Gemini 2.5 Flash5.57
4Voxtral-Mini-3B-25073.71
5Phi-4 Multimodal Instruct3.64
6Qwen3 Omni 30B A3B Instruct2.43

Interactive version: theaggregate.ai/benchmark?slug=sea-speechbench-temporal-localization-60-120-s · How It Works · Data refreshed daily, snapshot 2026-09-26.