LLaVA-NeXT-72B: benchmark results
Provider: LLaVA Team. Access: Open.
Aggregate Rating 1448 ± 35, rank #3863 of 10916 rated models, from 38 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MUSE - Activity Localization | 54.5 | Accuracy (%) | 79.3 |
| MUSE - Jigsaw Puzzle | 27 | Accuracy (%) | 72.4 |
| LifeEval (Multiple-Choice) - Safety and Feasibility Assessment | 73.78 | Accuracy (%) on the 328 multiple-choice Safety and Feasibili | 64 |
| LifeEval (Open-Ended) - Safety and Feasibility Assessment | 49.06 | GPT-5 judge score against the reference answer (0 to 1 in st | 64 |
| LifeEval (Multiple-Choice) | 66.26 | Accuracy (%) on LifeEval's 2,069 multiple-choice questions a | 60 |
| LifeEval (Multiple-Choice) - Contextual Knowledge Retrieval | 70.77 | Accuracy (%) on the 390 multiple-choice Contextual Knowledge | 60 |
| LifeEval (Multiple-Choice) - Dynamic Task Reasoning | 62.16 | Accuracy (%) on the 370 multiple-choice Dynamic Task Reasoni | 60 |
| LifeEval (Multiple-Choice) - Goal-oriented Planning | 64.19 | Accuracy (%) on the 377 multiple-choice Goal-oriented Planni | 60 |
| LifeEval (Multiple-Choice) - Multi-turn Interactive Collaboration | 73.2 | Accuracy (%) on the 269 multiple-choice Multi-turn Interacti | 60 |
| LifeEval (Open-Ended) - Goal-oriented Planning | 37.19 | GPT-5 judge score against the reference answer (0 to 1 in st | 60 |
| MUSE - Relative Position | 3 | Accuracy (%) | 56.9 |
| LifeEval (Open-Ended) | 38.24 | GPT-5 judge score against the reference answer (0 to 1 in st | 56 |
Interactive version: theaggregate.ai/model?slug=llava-next-72b · How It Works · Data refreshed daily, snapshot 2026-11-08.