Qwen2.5-Omni-3B: benchmark results
Provider: Alibaba. Released 2025-04-30. Access: Open.
Aggregate Rating 1472 ± 17, rank #2807 of 10916 rated models, from 164 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| HELM Robo-Reward-Bench - Droid Perspective | 0.06 | Absolute error | 100 |
| MCBench | 64.5 | Average accuracy (%) over the eight category-by-label cells | 100 |
| SEA-SpeechBench - Temporal Localization (30-60 s) | 18.21 | Span-overlap F1 (%; predict the start and end time at which | 91.7 |
| MERA Multi - AQUARIA | 49.5 | Score (%) | 90.9 |
| Omni-Persona | 36.7 | Calibrated accuracy (%): item-weighted accuracy over answera | 90 |
| DatBench - Document Understanding | 69.1 | Accuracy (%) | 88 |
| MERA Multi - Audio | 38 | Score (%) | 86.4 |
| MOV-Bench - Hypothetical Reasoning | 38.33 | Accuracy (%; 60 hypothetical-reasoning questions; direct inf | 85.7 |
| SEA-SpeechBench - Temporal Localization (0-30 s) | 30.49 | Span-overlap F1 (%; predict the start and end time at which | 85.7 |
| MERA Multi - ruEnvAQA | 50 | Score (%) | 84.1 |
| Omni-Persona - Audio-to-Audio | 25.9 | Calibrated accuracy (%): item-weighted accuracy over answera | 80 |
| Omni-Persona - Image-to-Image | 35.9 | Calibrated accuracy (%): item-weighted accuracy over answera | 80 |
Interactive version: theaggregate.ai/model?slug=qwen2-5-omni-3b · How It Works · Data refreshed daily, snapshot 2026-11-08.