LLaVA-NeXT-Video-7B-DPO: benchmark results
Provider: LLaVA Team. Access: Open.
Aggregate Rating 1342 ± 47, rank #7932 of 10916 rated models, from 43 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| TempCompass | 53.75 | Avg All (%) | 46.4 |
| VideoOdyssey-V - Summarization | 36.9 | Multiple-choice accuracy (%) on the 65 summarization questio | 38.2 |
| VideoOdyssey-V - Object Recognition | 30.8 | Multiple-choice accuracy (%) on the 208 object recognition q | 29.4 |
| VERHallu - Causal QA | 17.9 | Accuracy (%; 497 seven-option questions on why an event happ | 17.6 |
| VideoOdyssey-V - Counting | 18.2 | Multiple-choice accuracy (%) on the 307 counting questions ( | 17.6 |
| VideoOdyssey-V - Temporal Ordering | 23.6 | Multiple-choice accuracy (%) on the 89 temporal ordering que | 17.6 |
| VERHallu - Subevent QA | 16.7 | Accuracy (%; 258 seven-option questions on which event occur | 11.8 |
| VERHallu - Temporal QA | 17 | Accuracy (%; 212 seven-option questions on what happens afte | 11.8 |
| VideoOdyssey-V - Captioning | 30.6 | Multiple-choice accuracy (%) on the 49 captioning questions | 11.8 |
| VideoOdyssey-V - Emotional Reasoning | 28.2 | Multiple-choice accuracy (%) on the 71 emotional reasoning q | 11.8 |
| VideoOdyssey-V - Temporal Grounding | 24.6 | Multiple-choice accuracy (%) on the 65 temporal grounding qu | 11.8 |
| VERHallu - Counterfactual QA | 2.9 | Accuracy (%; 967 seven-option questions about an event absen | 5.9 |
Interactive version: theaggregate.ai/model?slug=llava-next-video-7b-dpo · How It Works · Data refreshed daily, snapshot 2026-11-08.