LLaVA-OneVision-1.5-4B: benchmark results
Provider: LLaVA Team.
Aggregate Rating 1538 ± 36, rank #1851 of 10916 rated models, from 72 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| LinkS2Bench - Spatiotemporal Ordering | 52.6 | Spatiotemporal Ordering: multiple-choice accuracy (%) on the | 100 |
| LinkS2Bench - Region Reachability | 63.7 | Region Reachability: yes/no accuracy (%) on whether a target | 91.2 |
| Sens-VisualNews | 86.7 | Top-1 accuracy (%) on the test split of the full dataset (ma | 81.8 |
| LinkS2Bench - Region Interaction | 47.5 | Region Interaction: multiple-choice accuracy (%) on whether | 52.9 |
| GST-Bench - Egocentric Direction (Semantic) | 17.63 | Mean accuracy (%; share of predicted egocentric directions t | 52.4 |
| GST-Bench - Global Position (Semantic) | 18.86 | Mean accuracy (%; predicted top-down map position of a previ | 52.4 |
| GST-Bench - Trajectory Selection | 25.59 | Accuracy (%; multiple choice of the camera trajectory drawn | 50 |
| GST-Bench - Top-Down Selection (Hard) | 27.53 | Accuracy (%; multiple choice of the top-down image of the ex | 47.6 |
| LinkS2Bench - Target Visibility | 64.7 | Target Visibility: yes/no accuracy (%) on whether a marked s | 47.1 |
| GST-Bench - Top-Down Selection (Medium) | 37.27 | Accuracy (%; multiple choice of the top-down image of the ex | 45.2 |
| EgoCoT-Bench - Location and Visibility Retrospection | 55.04 | Answer accuracy (%, four-way multiple choice, strict exact m | 41.7 |
| EgoCoT-Bench - State Retrospection | 61.51 | Answer accuracy (%, four-way multiple choice, strict exact m | 38.9 |
Interactive version: theaggregate.ai/model?slug=llava-onevision-1-5-4b · How It Works · Data refreshed daily, snapshot 2026-11-08.