VideoLLaMA2.1-7B-AV: benchmark results
Provider: Alibaba. Released 2024-10-22. Access: Open.
Aggregate Rating 1447 ± 36, rank #3901 of 10916 rated models, from 40 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OmniPro - Condition Alert | 24.1 | Probe-mode accuracy (%): for each ground-truth trigger the m | 100 |
| AffectSim (Active Observation) | 20.89 | Macro-F1 (%; A-Obs: the view reached by the authors' two-sta | 78.3 |
| AffectSim (Initial View) | 18.67 | Macro-F1 (%; P-Init: the agent's initial camera pose, held f | 78.3 |
| AffectSim (Monitor View) | 20.27 | Macro-F1 (%; P-Mon: a fixed room-level monitoring camera; fi | 73.9 |
| AffectSim (Reference View) | 27.65 | Macro-F1 (%; P-Ref: an informative camera trajectory chosen | 73.9 |
| OmniPro - Event Alert | 21.8 | Probe-mode accuracy (%): for each ground-truth trigger the m | 71.4 |
| TraceAV-Bench - Visual Counting | 38.94 | Accuracy (%) on the 226 Visual Counting (VC) questions, four | 64.3 |
| OmniCompass-AV - Spatial Grounding - Consistency Score | 8.6 | Consistency score (%) | 43.8 |
| OmniCompass-AV - Temporal Grounding - Consistency Score | 0.4 | Consistency score (%) | 21.9 |
| EmoBench-M | 48.7 | Average Score | 20 |
| OmniPro | 12.3 | Probe-mode accuracy (%): for each ground-truth trigger the m | 14.3 |
| OmniPro - Cumulative Counting | 4.1 | Probe-mode accuracy (%): for each ground-truth trigger the m | 14.3 |
Interactive version: theaggregate.ai/model?slug=videollama2-1-7b-av · How It Works · Data refreshed daily, snapshot 2026-11-08.