LLaVA-NeXT-13B: benchmark results

Provider: LLaVA Team.

Aggregate Rating 1403 ± 46, rank #6056 of 10489 rated models, from 27 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
CAGE - Counterfactual (Chain-Text Probe) - Chain Score7.61Judge score (0-10) for the causal chain in arrow notation, m100
CAGE - Counterfactual (Chain-Text Probe) - Text Score7.64Judge score (0-10) for the textual answer, mean of GPT-4o an100
CAGE - Intervention (Chain-Text Probe) - Chain Score7.83Judge score (0-10) for the causal chain in arrow notation, m100
CAGE - Intervention (Chain-Text Probe) - Text Score7.96Judge score (0-10) for the textual answer, mean of GPT-4o an100
EIBench (complex)39.16Recall91.7
EIBench54.33Recall83.3
PlantMicro - Host40.1Host plant species identification accuracy (%); four-option 80
PlantMicro - Localization mAP@0.535.5Mean average precision at IoU 0.5 (%) of directly predicted 75
CAGE - Counterfactual (Text-Only Probe)7.94Judge score (0-10), mean of GPT-4o and Claude 3.5 Sonnet rat71.4
CAGE - Intervention (Text-Only Probe)8.08Judge score (0-10), mean of GPT-4o and Claude 3.5 Sonnet rat71.4
CAGE - Association (Text-Only Probe)7.95Judge score (0-10), mean of GPT-4o and Claude 3.5 Sonnet rat57.1
PlantMicro - Pathogen29.56Pathogen identification accuracy (%); four-option multiple c50

Interactive version: theaggregate.ai/model?slug=llava-next-13b · How It Works · Data refreshed daily, snapshot 2026-11-20.