Daily-Omni
Multimodal · 2025-05-23
Daily-Omni is an audio-visual question-answering benchmark of 1,197 multiple-choice questions on 684 real-world videos (647 on 30-second and 550 on 60-second clips) across six task families that require cross-modal temporal reasoning. Models receive the video with its soundtrack, and scores are overall multiple-choice accuracy against a stated 25% random-guess baseline.
Top models (higher is better)
| Model | Score |
|---|---|
| LFM3-Omni-2B | 70.1 |
| Phi-4-multimodal-instruct | 48.9 |