OmniBench (speech audio)
Multimodal · 2024-09-23
The 771-question subset of OmniBench whose audio clip is human speech, scored as four-option accuracy with the official parser. It comes from the same runs as the full OmniBench score, so it is a component of that family rather than independent evidence.
Top models (higher is better)
| Model | Score |
|---|---|
| LFM3-Omni-2B | 48.3 |
| Phi-4-multimodal-instruct | 38.9 |