SEA-SpeechBench - Temporal Localization (0-30 s): leaderboard

Metric: Span-overlap F1 (%; predict the start and end time at which a queried phrase is spoken in a recording of 0-30 seconds; English instruction prompt; blank where the model cannot take audio of that length). Source: arxiv.org. Saturation forecast: Around 2035. 15 models tracked.

Top models

#ModelScore
1Qwen2.5-Omni-7B35.74
2Qwen2-Audio-7B-Instruct33.3
3GPT-4o Audio26.47
4Voxtral-Mini-3B-250717.85
5Gemma 3n E4B (IT)13.25
6Phi-4 Multimodal Instruct12.97
7gemma-3n-E2B-it11.82
8Gemini 2.5 Flash11.66
9SeaLLMs-Audio-7B11.57
10Qwen3 Omni 30B A3B Instruct11.4

Interactive version: theaggregate.ai/benchmark?slug=sea-speechbench-temporal-localization-0-30-s · How It Works · Data refreshed daily, snapshot 2026-09-26.