Benchmark
JointAVBench
What is JointAVBench?
An ICLR 2026 poster on joint audio-visual reasoning in Omni-LLMs, built with strict audio-video correlation across five cognitive dimensions and four audio…
- Released
- 2025-12-14
- Evaluates
- multimodal
- Openness
- unknown
- Importer
- Model reports
- Reported scores
- 17
Source provenance
- Original evidence https://arxiv.org/abs/2512.12772
JointAVBench paper
JointAVBench results and reported scores
Model reports
| Model | Organization | Reported value | Reported | Evidence |
|---|---|---|---|---|
| AVicuna | University of Rochester | 31.0 | 2026-05-14 | evidence |
| Gemini 2.5 Flash | 58.0 | 2026-05-14 | evidence | |
| Gemini 2.5 Pro | 65.3 | 2026-05-14 | evidence | |
| GPT-4o | OpenAI | 45.0 | 2026-05-14 | evidence |
| InternVL-2.5 | OpenGVLab | 51.7 | 2026-05-14 | evidence |
| Kimi-Audio | Moonshot AI | 45.6 | 2026-05-14 | evidence |
| LLaVA-Video | ByteDance | 47.1 | 2026-05-14 | evidence |
| OneLLM | Tencent | 36.9 | 2026-05-14 | evidence |
| Qwen2-Audio | Alibaba | 39.5 | 2026-05-14 | evidence |
| Qwen2.5-Omni | Alibaba | 56.5 | 2026-05-14 | evidence |
| Qwen2.5-VL | Alibaba | 47.7 | 2026-05-14 | evidence |
| Qwen3-Omni | Alibaba | 63.6 | 2026-05-14 | evidence |
| v-SALMONN | Shanghai Jiao Tong University | 35.8 | 2026-05-14 | evidence |
| v-SALMONN-2+ | Shanghai Jiao Tong University | 46.7 | 2026-05-14 | evidence |
| v-SALMONN-o1 | Shanghai Jiao Tong University | 36.4 | 2026-05-14 | evidence |
| VideoLLaMA2 | Alibaba | 46.8 | 2026-05-14 | evidence |
| VideoLLaMA3 | Alibaba | 50.0 | 2026-05-14 | evidence |
Scores are partitioned by the source that reported them and are never merged into a single cross-source ranking, because the sources measure different things and say so.
Which sources cite JointAVBench?
- JointAVBench v2 results table Renmin University of China