InstructBLIP-Vicuna-7B: benchmark results

Provider: Salesforce. Released 2023-05-11. Access: Open.

Aggregate Rating 1273 ± 47, rank #8898 of 10489 rated models, from 37 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
CoreCognition Image only Conservation65.4Accuracy (%)86.8
MMBench-Live - Attribute Reasoning91.73Accuracy (%; attribute reasoning questions; 5.9K newly gener80
ProactiveBench - QuickDraw18.3Trajectory accuracy (%) on the partial QuickDraw sketches: m76.2
ProactiveBench - ChangeIt24.5Trajectory accuracy (%) on the ambiguous ChangeIt video fram61.9
MMBench-Live - Coarse Perception92.05Accuracy (%; coarse perception questions; 5.9K newly generat60
ProactiveBench - MVP-N8.8Trajectory accuracy (%) on the uninformative multi-view obje57.1
ChartBench - Chart Recognition49.57Acc+ (%)55
ProactiveBench - COCO12.6Trajectory accuracy (%) on the MS-COCO scenes that need a ca47.6
ProactiveBench - VSOD9.5Trajectory accuracy (%) on the temporally occluded video fra33.3
ProactiveBench12.2Trajectory accuracy (%) averaged over the seven scenarios (R28.6
CoreCognition Image only Spatiality42.99Accuracy (%)22.4
GMAI-MMBench30.95Accuracy on single-choice questions (%)21.7

Interactive version: theaggregate.ai/model?slug=instructblip-vicuna-7b · How It Works · Data refreshed daily, snapshot 2026-11-20.