Qwen3 Omni Flash: benchmark results
Provider: Alibaba.
Aggregate Rating 1538 ± 49, rank #1852 of 10916 rated models, from 9 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| EmoS | 58 | Accuracy (%; Basic-7 emotion label (anger, disgust, fear, jo | 66.7 |
| EcoG-Bench (Video-Omni) - Referent Classification | 47.1 | Classification accuracy (%): share of referents for which th | 60 |
| EcoG-Bench (Video-Omni) | 0.7 | Eco-accuracy (%): share of referents whose option (what), 2D | 50 |
| AGL1K | 1498.92 | Geo-score (0-5000) | 46.7 |
| AGL1K - Non-Speech Distance Error | 6235.23 | Mean distance error on clips without speech (km) | 46.7 |
| EcoG-Bench (Video-Omni) - Sequence Success | 0.6 | Sequence success (%): share of clips in which every referent | 40 |
| AA Speech to Speech - Conversational Dynamics (Full Duplex Bench) | 72.65 | Score (0-100), AA's weighted Full Duplex Bench v1/v1.5 subse | 22.9 |
| AGL1K - Speech Distance Error | 4741.73 | Mean distance error on clips with speech (km) | 20 |
| AA Speech to Speech - Big Bench Audio | 58.7 | Accuracy (%) | 18.6 |
Interactive version: theaggregate.ai/model?slug=qwen3-omni-flash · How It Works · Data refreshed daily, snapshot 2026-11-08.