Long-Form Speech Understanding on AudioMarathon 1.0 (test)
65.6Average ScoreQwen2.5-Omni-7B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Qwen2.5-Omni-7BParameters=7B2026.02 | 65.6 | 70.5 | |
| Qwen2.5-Omni-3BParameters=3B2026.02 | 62.9 | 67.2 | |
| Audio-Flamingo-32026.02 | 60.6 | 63 | |
| Gemini-2.5-Flash2026.02 | 59.6 | 59.6 | |
| Voxtral-Mini-3B-2507Parameters=3B, Version=25072026.02 | 57 | 57.4 | |
| Upper-bound2026.02 | 56.1 | 51.6 | |
| Speech-XL2026.02 | 54.7 | 48.9 | |
| Phi-4-Multimodal2026.02 | 51.5 | 47.7 | |
| Gemini-2.5-Flash-Lite2026.02 | 50.1 | 50.6 | |
| Gemma-3n-E4B-itModel Size=4B2026.02 | 49.4 | 49.3 | |
| GPT-4o-Audio (2024-12-17)Release Date=2024-12-172026.02 | 48.3 | 48.7 | |
| GPT-4o-Audio (2024-10-01)Release Date=2024-10-012026.02 | 48 | 47.4 | |
| Gemma-3n-E2B-itModel Size=2B2026.02 | 46.8 | 45.5 | |
| Baichuan-Omni-1.52026.02 | 38.4 | 39.3 | |
| Aero-1-Audio2026.02 | 36.1 | 42.8 | |
| Audio-Flamingo-22026.02 | 31.8 | 35.6 |