VILA-1.5-8B: benchmark results
Provider: NVIDIA. Released 2024-05-06. Access: Open.
Aggregate Rating 1427 ± 46, rank #4802 of 10916 rated models, from 18 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| VLM-3R Release VSI-Bench - Obj. Size | 50.3 | Mean relative accuracy (%) | 66.7 |
| CG-Bench - Accuracy@IoU | 1.43 | Accuracy@IoU (%) | 52.4 |
| CG-Bench - Open-Ended | 23.8 | Open-Ended Accuracy (%) | 38.1 |
| CG-Bench | 28.1 | Long-Video MCQ Accuracy (%) | 33.3 |
| CG-Bench - Clue Grounding mIoU | 1.98 | Temporal Grounding mIoU (%) | 33.3 |
| CG-Bench - Clue MCQ | 41.4 | Clue-Interval MCQ Accuracy (%) | 33.3 |
| VLM-3R Release VSI-Bench - Abs. Dist. | 21.8 | Mean relative accuracy (%) | 33.3 |
| VLM-3R Release VSTI-Bench - Cam. Mov. Dir. | 42.2 | Accuracy (%) | 33.3 |
| VLM-3R Release VSI-Bench - Appr. Order | 24.8 | Accuracy (%) | 26.7 |
| VLM-3R Release VSI-Bench - Rel. Dir. | 34.8 | Accuracy (%) | 26.7 |
| VLM-3R Release VSI-Bench - Route Plan | 31 | Accuracy (%) | 26.7 |
| VLM-3R Release VSI-Bench - Overall | 28.9 | Source benchmark score (%) | 20 |
Interactive version: theaggregate.ai/model?slug=vila-1-5-8b · How It Works · Data refreshed daily, snapshot 2026-11-08.