InstructBLIP-7B: benchmark results

Provider: Salesforce. Released 2023-06-13. Access: Open.

Aggregate Rating 1282 ± 42, rank #8811 of 10489 rated models, from 414 benchmark results.

Strongest benchmark results

BenchmarkScoreMetricPercentile
OpenVLM MMT-Bench - Image Matting40Score (%)97.8
OpenVLM MMT-Bench - Gaze Estimation35Score (%)93
OpenVLM COCO Captions58.1CIDEr91.9
OpenVLM MMT-Bench - Action Quality Assessment30Score (%)84.7
OpenVLM MMT-Bench - Multiple View Image Understanding35Score (%)82.3
OpenVLM MMT-Bench - Order Hallucination50Score (%)81.8
OpenVLM MMT-Bench - Artwork Emotion Recognition55Score (%)74.8
OpenVLM MMT-Bench - 3D Indoor Recognition35Score (%)68.2
OpenVLM MMT-Bench - Fashion Recognition75Score (%)67.7
OpenVLM MMT-Bench - Depth Estimation55Score (%)67.5
OpenVLM MMT-Bench - Handwritten Retrieval30Score (%)67.2
OpenVLM MMT-Bench - 3D47.5Score (%)64.6

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