SEA-SpeechBench - Temporal Localization (30-60 s): leaderboard

Metric: Span-overlap F1 (%; predict the start and end time at which a queried phrase is spoken in a recording of 30-60 seconds; English instruction prompt; blank where the model cannot take audio of that length). Source: arxiv.org. Saturation forecast: Around 2034. 13 models tracked.

Top models

#ModelScore
1Qwen2.5-Omni-7B19.98
2GPT-4o Audio17.83
3Voxtral-Mini-3B-25079.87
4Gemma 3n E4B (IT)8.72
5gemma-3n-E2B-it8.27
6Gemini 2.5 Flash7.7
7Phi-4 Multimodal Instruct6.25
8Qwen3 Omni 30B A3B Instruct5.9

Interactive version: theaggregate.ai/benchmark?slug=sea-speechbench-temporal-localization-30-60-s · How It Works · Data refreshed daily, snapshot 2026-09-26.