LLaVA-Video-7B: benchmark results
Provider: LLaVA Team. Released 2024-10-04. Access: Open.
Aggregate Rating 1510 ± 19, rank #2206 of 10916 rated models, from 80 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Cambrian-S Release - Percept. Test † | 67.9 | Accuracy (%) | 90 |
| TOC-Bench - Event Count | 26.3 | Event Count accuracy (%) on TOC-Bench (2,323 human-verified | 90 |
| TOC-Bench - Reappearance Identity | 100 | Reappearance Identity accuracy (%) on TOC-Bench (2,323 human | 90 |
| Cambrian-S Release - CV-Bench | 75.7 | Accuracy (%) | 83.3 |
| SFI-Bench - Global and Conditional Counting | 55.4 | Accuracy (%) on the global and conditional counting question | 83.3 |
| Flat-Pack Bench - Part Mating | 52.87 | Mate: decide whether two parts are connected in the final as | 81 |
| VIABench - VQA | 39 | Visual question answering score (0-100): answers to question | 71.4 |
| Cambrian-S Release - LongVBench † | 58.2 | Accuracy (%) | 66.7 |
| Cambrian-S Release - Video MME † | 63.3 | Accuracy (%) | 66.7 |
| VLM2-Bench | 43.32 | Average Score on VLM2-bench (9 subtasks) | 62.5 |
| RealStreamEval - Forward Active Responding | 15.3 | Score (%; RealStreamEval on OVO-Bench tasks and questions: f | 60 |
| Cambrian-S Release - VSI-Bench | 35.6 | Source benchmark score (%) | 52.6 |
Interactive version: theaggregate.ai/model?slug=llava-video-7b · How It Works · Data refreshed daily, snapshot 2026-11-08.