LLaVA-OneVision: benchmark results
Provider: LLaVA Team.
Aggregate Rating 1526 ± 33, rank #1994 of 10916 rated models, from 24 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| UniG2U-Bench (Direct) - Physics Reasoning | 44 | Accuracy (%) on the 200 mechanics and optics items, mean of | 100 |
| UniG2U-Bench (Direct) - Real-world Applications | 43 | Accuracy (%) on the 200 real-world application items (attent | 88.5 |
| UniG2U-Bench (Direct) | 33.35 | Accuracy (%) over the 3,000 UniG2U-Bench image questions (20 | 80.8 |
| UniG2U-Bench (Direct) - Puzzles and Games | 24.48 | Accuracy (%) on the 537 puzzle and game items (mental recons | 80.8 |
| UniG2U-Bench (Direct) - Perception Reasoning | 37.13 | Accuracy (%) on the 1,263 perception items (IllusionBench sc | 76.9 |
| Video-MME - Long Videos (w/o subs) | 60 | Accuracy (%) | 76 |
| Video-MME - Short Videos (w/ subs) | 79.3 | Accuracy (%) | 70 |
| UniG2U-Bench (Direct) - Spatial Intelligence | 25.4 | Accuracy (%) on the 500 MMSI-Bench spatial items (multi-step | 65.4 |
| Video-MME - Short Videos (w/o subs) | 76.7 | Accuracy (%) | 62 |
| UniG2U-Bench (Direct) - Geometry Reasoning | 10.5 | Accuracy (%) on the 200 planar and solid geometry items, mea | 55.8 |
| AICA-Bench - Emotion Understanding (Basic) | 54.02 | Weighted F1 (%) of emotion classification (expressed emotion | 54.5 |
| UniG2U-Bench (Direct) - Chart Reasoning | 78 | Accuracy (%) on the 100 ChartQA items, from UniG2U-Bench, an | 53.8 |
Interactive version: theaggregate.ai/model?slug=llava-onevision · How It Works · Data refreshed daily, snapshot 2026-11-08.