Audio-Visual Understanding on AVHBench
81.7Overall ScoreTAC-V
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| TAC-VArchitecture=Describer + Text-only Reasoner, Reasoner=Gemini32026.02 | 81.7 | — | — | — | — | |
| TAC-VArchitecture=Describer + Text-only Reasoner, Reasoner=Qwen32026.02 | 79.8 | — | — | — | — | |
| TAC-VArchitecture=Describer + Text-only Reasoner, Reasoner=Gemini32026.02 | 76.6 | — | — | — | — | |
| TAC-VArchitecture=Describer + Text-only Reasoner, Reasoner=Qwen32026.02 | 76.1 | — | — | — | — | |
| VLMArchitecture=Describer + Text-only Reasoner, Reasoner=Qwen32026.02 | 70.8 | — | — | — | — | |
| PandaGPTArchitecture=Native MLLM2026.02 | 61.3 | — | — | — | — | |
| PandaGPTArchitecture=Native MLLM2026.02 | 58.5 | — | — | — | — | |
| VLMArchitecture=Describer + Text-only Reasoner, Reasoner=Qwen32026.02 | 51.8 | — | — | — | — | |
| AVLM-2BModel Scale=2B2026.06 | — | — | — | — | 61.6 | |
| Gemma-4-E2B-itModel Scale=4B2026.06 | — | — | — | — | 50.72 | |
| Qwen2.5-Omni-3BModel Scale=3B2026.06 | — | — | — | — | 71 | |
| video-SALMONN2FLOPs=100, Latency=0.44, Memory=28G2026.01 | — | 64.5 | 50.8 | 3.2 | — | |
| video-SALMONN2 w/ FastAVFLOPs=58, Latency=0.29, Memory=21G2026.01 | — | 64.8 | 50.7 | 3.1 | — | |
| VideoLLaMA2FLOPs=100, Latency=0.43, Memory=22G2026.01 | — | 77.9 | 57.8 | 2.8 | — | |
| VideoLLaMA2 w/ FastAVFLOPs=56, Latency=0.32, Memory=19G2026.01 | — | 78.2 | 69 | 2.9 | — |