LLaVA-1.5-7B: benchmark results
Provider: LLaVA Team. Released 2023-10-05. Access: Open.
Aggregate Rating 1393 ± 18, rank #6413 of 10916 rated models, from 284 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| ProactiveBench - ChangeIt | 44.2 | Trajectory accuracy (%) on the ambiguous ChangeIt video fram | 100 |
| SiT-Bench - Global Perception & Mapping | 29.18 | Accuracy (%; item-weighted over the category's subtasks) | 96.7 |
| ProactiveBench - QuickDraw | 25.5 | Trajectory accuracy (%) on the partial QuickDraw sketches: m | 95.2 |
| EgoPoint-Bench - Real-World Adversarial Resilience | 69.37 | Accuracy (%) on the Adversarial Resilience questions among t | 90.9 |
| ProactiveBench - VSOD | 26.2 | Trajectory accuracy (%) on the temporally occluded video fra | 90.5 |
| UAVBench - VQA Comparison | 72.21 | Image VQA accuracy (%) on comparison questions, zero-shot wi | 90 |
| UAVBench - VQA Weather | 88.98 | Image VQA accuracy (%) on weather-condition recognition, zer | 90 |
| UAVBench - Region Classification (DroneVehicle) | 76.78 | Region-level classification accuracy (%) of a boxed object, | 87.5 |
| ProactiveBench - COCO | 32.3 | Trajectory accuracy (%) on the MS-COCO scenes that need a ca | 85.7 |
| PathBind - Text-only VQA D3 | 40 | Score (%) | 82.4 |
| UAVBench - Region Classification (WebUAV-3M) | 15.64 | Region-level classification accuracy (%) of a boxed object, | 80 |
| ProactiveBench | 24.8 | Trajectory accuracy (%) averaged over the seven scenarios (R | 76.2 |
Interactive version: theaggregate.ai/model?slug=llava-1-5-7b · How It Works · Data refreshed daily, snapshot 2026-11-08.