LongVA-7B: benchmark results
Provider: LMMs-Lab. Released 2024-06-24. Access: Open.
Aggregate Rating 1480 ± 27, rank #2664 of 10916 rated models, from 65 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MLVU-Dev - Generation | 4.33 | Generation score, GPT-4 judged (0-10) | 86.2 |
| VG-LLM Release v3 - VSIRelDirection | 43.3 | Accuracy (%) | 83.3 |
| VLM-3R Release VSI-Bench - Rel. Dir. | 43.3 | Accuracy (%) | 80 |
| MLVU-Test - Sports QA | 44.4 | Multiple-choice accuracy (%) | 76.5 |
| MLVU-Test - Needle QA | 46.7 | Multiple-choice accuracy (%) | 72.1 |
| MLVU-Test - Generation | 3.91 | Generation score, GPT-4 judged (0-10) | 71.9 |
| MLVU-Test - Topic Reasoning | 81.3 | Multiple-choice accuracy (%) | 70.6 |
| MLVU-Test - Action Count | 23.3 | Multiple-choice accuracy (%) | 69.1 |
| DriveSpatial - Understanding | 35.9 | Accuracy (%) on the multiple-choice spatial-understanding ta | 63.6 |
| MLVU-Test - Ego Reasoning | 39.6 | Multiple-choice accuracy (%) | 61.8 |
| MLVU-Test - Plot QA | 46 | Multiple-choice accuracy (%) | 61.8 |
| MLVU-Dev | 56.3 | Multiple-choice accuracy (%) | 60.5 |
Interactive version: theaggregate.ai/model?slug=longva-7b · How It Works · Data refreshed daily, snapshot 2026-11-08.