LLaVA-OneVision-1.5-4B-Instruct: benchmark results
Provider: LLaVA Team. Access: Open.
Aggregate Rating 1558 ± 47, rank #1587 of 10916 rated models, from 18 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| CVSBench - CVUSA Ground-to-Satellite | 68.98 | Accuracy (%) on CVUSA-subset ground-to-satellite questions: | 78.6 |
| CVSBench - CVUSA Satellite-to-Ground | 60.09 | Accuracy (%) on CVUSA-subset satellite-to-ground questions: | 69.2 |
| NextMotionQA - Question Answering | 51.32 | Task 1 Jaccard (%) between predicted and gold option subsets | 54.5 |
| OddGridBench - Size | 3 | Accuracy (%) on the 200 images in which the odd item differs | 44.4 |
| OddGridBench - Color | 27.5 | Accuracy (%) on the 200 images in which the odd item differs | 38.9 |
| OddGridBench - Rotation | 7.5 | Accuracy (%) on the 200 images in which the odd item differs | 38.9 |
| NextMotionQA - Error Correction | 46.16 | Task 3 caption error correction score (0-100): mean of error | 36.4 |
| CVSBench - Viewpoint Arrow | 27.95 | Accuracy (%) on View-Arrow: choose the camera-pose arrow on | 35.7 |
| EscherVerse | 43.75 | Overall Acc. (%) | 34.6 |
| OddGridBench | 13.79 | Accuracy (%) over all 1,400 OddGridBench synthetic grid imag | 33.3 |
| OddGridBench - Four Attributes | 26 | Accuracy (%) on the 200 images in which the odd item differs | 33.3 |
| OddGridBench - Two Attributes | 14 | Accuracy (%) on the 200 images in which the odd item differs | 33.3 |
Interactive version: theaggregate.ai/model?slug=llava-onevision-1-5-4b-instruct · How It Works · Data refreshed daily, snapshot 2026-11-08.