Qwen3-Omni-30B-A3B-Captioner: benchmark results
Provider: Alibaba. Released 2025-09-22. Access: Open.
Aggregate Rating 1615 ± 30, rank #880 of 10916 rated models, from 18 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MOV-Bench - Causal Reasoning | 59.34 | Accuracy (%; 91 causal-reasoning questions; direct inference | 100 |
| MOV-Bench | 53.37 | Accuracy (%; 519 audio-visual multi-hop multiple-choice ques | 85.7 |
| MOV-Bench - Intent Reasoning | 57.5 | Accuracy (%; 80 intent-reasoning questions; direct inference | 85.7 |
| AnyAudio-Judge Bench - Chinese (Holistic) | 65.33 | Classification accuracy (%) on the Chinese instruction subse | 83.3 |
| AnyAudio-Judge Bench - English (Holistic) | 64.24 | Classification accuracy (%) on the English instruction subse | 83.3 |
| UGC-VideoCap - Average | 72.5 | Mean judge score (audio, visual, detail) | 75 |
| MOV-Bench - Referential Reasoning | 57.76 | Accuracy (%; 161 referential-reasoning questions; direct inf | 71.4 |
| DiaDemBench - Speaker attribution F1 (REF) | 43.9 | REF F1 | 69.2 |
| DiaDemBench - Utterance transcription F1 (ASR) | 58.8 | ASR F1 | 69.2 |
| CapRiCorn-1K | 14.3 | Keypoint accuracy (%; share of annotated keypoints a GPT-4.1 | 64.7 |
| Daily-Omni (caption QA, Gemini-2.5-Pro reader) | 27.2 | Accuracy (%) | 60 |
| WorldSense (caption QA, Gemini-2.5-Pro reader) | 14.1 | Accuracy (%) | 60 |
Interactive version: theaggregate.ai/model?slug=qwen3-omni-30b-a3b-captioner · How It Works · Data refreshed daily, snapshot 2026-11-08.