Audio Event Grounding on AEGBench (test)
48mIoUQwen3-Omni-30B
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Qwen3-Omni-30Btraining_stage=SFT + GRPO, parameters=30B2026.07 | 48 | 52.4 | 69.7 | 55.9 | 56.6 | 46.5 | |
| Qwen2.5-Omni-7Btraining_stage=SFT, parameters=7B2026.07 | 42.4 | 41.6 | 65.7 | 41.1 | 50.8 | 40.2 | |
| Qwen2.5-Omni-7Btraining_stage=SFT + GRPO, parameters=7B2026.07 | 39.9 | 47.4 | 60.9 | 59.4 | 43.5 | 38.4 | |
| Qwen3-Omni-30Btraining_stage=SFT, parameters=30B2026.07 | 37.1 | 34.3 | 64.2 | 36.8 | 38.7 | 31.8 | |
| Qwen2.5-Omni-7Btraining_stage=Zero-shot, parameters=7B2026.07 | 32.4 | 34 | 60.2 | 33.1 | 49 | 26.4 | |
| Gemini-3-Promode=External zero-shot2026.07 | 32.3 | 28.2 | 57.4 | 41.3 | 32 | 28.9 | |
| Qwen3-Omni-30Btraining_stage=Zero-shot, parameters=30B2026.07 | 27.6 | 29.5 | 52.8 | 46.3 | 33.3 | 24.3 | |
| Qwen2-Audio-7Bmode=External zero-shot, parameters=7B2026.07 | 15.7 | 18.7 | 34.8 | 53.8 | 28.1 | 8.3 | |
| Kimi-Audio-7Bmode=External zero-shot, parameters=7B2026.07 | 11.7 | 16 | 28 | 19.3 | 31.4 | 7 | |
| Audio Flamingo Nextmode=External zero-shot2026.07 | 2.8 | 2.7 | 6.4 | 91.9 | 4.6 | 2 |