SEA-SpeechBench - Temporal Localization (120-180 s): leaderboard

Metric: Span-overlap F1 (%; predict the start and end time at which a queried phrase is spoken in a recording of 120-180 seconds; English instruction prompt; blank where the model cannot take audio of that length). Source: arxiv.org. Saturation forecast: Around 2030. 7 models tracked.

Top models

#ModelScore
1GPT-4o Audio5.38
2Gemini 2.5 Flash5.3
3Voxtral-Mini-3B-25072.45
4Qwen3 Omni 30B A3B Instruct1.28

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