Video-LLaVA: benchmark results
Provider: Other. Access: Open.
Aggregate Rating 1393 ± 30, rank #6421 of 10916 rated models, from 93 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| OpenVLM Video - MVBench - Object Shuffle | 42.5 | Accuracy (%) | 84.7 |
| VidPair-Halluc | 31.9 | Binary-QA weighted accuracy wAcc (%): question-pair and vide | 71.4 |
| OpenVLM Video - MLVU - Action Count | 34.5 | Accuracy (%) | 65.3 |
| EIBench (complex) | 30.9 | Recall | 50 |
| FineDiving-QA (Video-STaR) - Answer Score | 2.2 | Mean GPT-3.5-Turbo answer rating (1-5) on held-out FineDivin | 50 |
| VidPair-Halluc - Multi-Choice F1 | 53.52 | Multiple-choice F1 (%) for assessing and ranking clip-level | 50 |
| OpenVLM Video - MMBench-Video - Hallucination | 0.42 | GPT-4 Rating (0-3) | 42 |
| EIBench | 49.26 | Recall | 41.7 |
| OpenVLM Video - MVBench - Action Count | 43.5 | Accuracy (%) | 41.7 |
| DAR (Dynamic Affective Reasoning) - Affective Segmentation | 32.6 | Mean temporal IoU (%; overlap between predicted and ground-t | 36.4 |
| DAR (Dynamic Affective Reasoning) - Emotion Accuracy | 7 | Emotion accuracy (%; a segment counts only if its predicted | 36.4 |
| DAR (Dynamic Affective Reasoning) - Reasoning Quality | 1.7 | GPT-Score (out of 5): GPT-4o judge rating of each rationale | 36.4 |
Interactive version: theaggregate.ai/model?slug=video-llava · How It Works · Data refreshed daily, snapshot 2026-11-08.