LLaVA-NeXT-110B: benchmark results

Provider: LLaVA Team. Access: Open.

Aggregate Rating 1522 ± 37, rank #1954 of 10552 rated models, from 27 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
CoreCognition Image + Video Intentionality82.87Accuracy (%)97.5
CoreCognition Image + Video Hierarchy65.94Accuracy (%)81.9
CoreCognition Image + Video Mechanicalreasoning66.67Accuracy (%)81.5
CoreCognition Image + Video Permanence45.1Accuracy (%)73.9
LifeEval (Multiple-Choice) - Safety and Feasibility Assessment76.22Accuracy (%) on the 328 multiple-choice Safety and Feasibili72
CoreCognition Image + Video Intuitivephysics57.27Accuracy (%)70.2
LifeEval (Open-Ended) - Safety and Feasibility Assessment48.19GPT-5 judge score against the reference answer (0 to 1 in st60
CoreCognition Image + Video Mean54.88Accuracy (%)56.3
LifeEval (Multiple-Choice)65.45Accuracy (%) on LifeEval's 2,069 multiple-choice questions a56
LifeEval (Multiple-Choice) - Contextual Knowledge Retrieval70.26Accuracy (%) on the 390 multiple-choice Contextual Knowledge56
LifeEval (Multiple-Choice) - Dynamic Task Reasoning60.81Accuracy (%) on the 370 multiple-choice Dynamic Task Reasoni56
LifeEval (Multiple-Choice) - Goal-oriented Planning59.95Accuracy (%) on the 377 multiple-choice Goal-oriented Planni56

Interactive version: theaggregate.ai/model?slug=llava-next-110b · How It Works · Data refreshed daily, snapshot 2026-11-16.