UniSense Study - Audio-Image - TextVQA s: leaderboard
UniSense (2026): Audio-Image-text evaluation from the main paper table. Video-MME has no subtitles; audio-image variants provide spoken questions. Different input modes are separate benchmarks; source-reported comparison results retain their original identities. Score (%); higher is better.
Metric: Score (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | UniSense-9B | 80.4 | |
| 2 | Lyra-Base-9B | 80 | |
| 3 | UniSense-2B | 73.8 | |
| 4 | Lyra-Mini-3B | 73.4 | |
| 5 | Qwen3-Omni-30B-A3B | 70.8 | |
| 6 | Intern-Omni | 69.1 |
Interactive version: theaggregate.ai/benchmark?slug=unisense-study-audio-image-textvqa-s · How It Works · Data refreshed daily, snapshot 2026-10-21.