VideoLLaMA3-2B: benchmark results
Provider: Alibaba. Released 2025-01-21.
Aggregate Rating 1487 ± 40, rank #2554 of 10916 rated models, from 8 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MA-Bench - Fine-Grained Recognition | 29.6 | Accuracy (%) on the fine-grained (action-level) micro-action | 61.9 |
| MA-Bench (Closed-Ended) | 43.2 | Accuracy (%) averaged over the six closed-ended tasks (coars | 52.4 |
| MA-Bench - Coarse-Grained Recognition | 18.6 | Accuracy (%) on the coarse-grained (body-level) micro-action | 47.6 |
| CCTVBench | 6.48 | Quadruple accuracy (%): share of question pairs in which the | 44.4 |
| MMVU | 39.6 | Hidden-test overall accuracy, chain-of-thought (%) | 37.1 |
| MA-Bench (Open-Ended) | 0.12 | GPT-4o judge score (0 to 5) averaged over the six open-ended | 33.3 |
| FAVOR-Bench | 32.98 | Close-ended ALL accuracy (%) | 30 |
| CCTVBench - Balanced Accuracy | 55.38 | Balanced accuracy (%) over all binary video-question instanc | 11.1 |
Interactive version: theaggregate.ai/model?slug=videollama3-2b · How It Works · Data refreshed daily, snapshot 2026-11-08.