LLaVA-NeXT-Video-7B: benchmark results
Provider: LLaVA Team. Access: Open.
Aggregate Rating 1444 ± 19, rank #4020 of 10916 rated models, from 157 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| CrashSight - Scene Identification | 75.5 | Accuracy (%) on scene identification questions (road type, l | 100 |
| LongVQUBench - GQU Trend Assessment | 58.3 | Accuracy (%; Trend assessment (global quality understanding) | 100 |
| SayNext-Bench (PC19K) - BLEU-4 | 0.46 | Corpus BLEU-4 (%, 0-100) of the predicted against the transc | 100 |
| SayNext-Bench (PC19K) - ROUGE-L | 13.79 | Mean ROUGE-L F1 (%, 0-100) of the predicted against the tran | 100 |
| VLM-3R Release VSTI-Bench - Cam. Mov. Dir. | 49.8 | Accuracy (%) | 91.7 |
| VLM-3R Release VSI-Bench - Rel. Dist. | 43.5 | Accuracy (%) | 86.7 |
| CrashSight - Involved Parties | 52.3 | Accuracy (%) on involved-party questions (vehicle and vulner | 80 |
| VLM-3R Release VSI-Bench - Obj. Count | 48.5 | Mean relative accuracy (%) | 73.3 |
| VLM-3R Release VSI-Bench - Rel. Dir. | 42.4 | Accuracy (%) | 73.3 |
| VLM-3R Release VSI-Bench - Route Plan | 34 | Accuracy (%) | 70 |
| EgoMonth - Personality Inference | 62.5 | Accuracy (%; 16 personality inference questions; four-option | 68.2 |
| SayNext-Bench (PC19K) - BERTScore | 52.9 | BERTScore F1 with deberta-xlarge-mnli (times 100, 0-100) of | 66.7 |
Interactive version: theaggregate.ai/model?slug=llava-next-video-7b · How It Works · Data refreshed daily, snapshot 2026-11-08.