List your company
ResearchFormerly WizWand

Multi-image reasoning on QBench2 (val)

79.3AccuracyMM1.5-30B
39.88450.11760.3570.583Sep 30, 2024
Updated 8mo ago

Evaluation results

MethodLinks
MM1.5-30BModel Scale=>14B2024.0979.3
LLaVA-NeXT-Interleave-14BModel Scale=>14B2024.0976.7
GPT-4VModel Scale=>14B2024.0976.5
Mantis-Idefics2-8BModel Scale=7B2024.0975.2
BLIP-3-4BModel Scale=3B2024.0975.1
LLaVA-NeXT-Interleave-7BModel Scale=7B2024.0974.2
MM1.5-3B-MoEModel Scale=3B2024.0973.8
MM1.5-3BModel Scale=3B2024.0973.2
MM1.5-7BModel Scale=7B2024.0973.2
MM1.5-1B-MoEModel Scale=1B2024.0970.9
MM1.5-1BModel Scale=1B2024.0966.4
Idefics2-8BModel Scale=7B2024.0957
Phi-3-Vision-4BModel Scale=3B2024.0956.8
LLaVA-NeXT-Interleave-0.5BModel Scale=1B2024.0952
Emu2-Chat-37BModel Scale=>14B2024.0950.1
LLaVA-v1.5-7BModel Scale=7B2024.0949.3
LLaVAOneVision-0.5BModel Scale=1B2024.0948.8
MM1-7BModel Scale=7B2024.0943.6
MM1-1BModel Scale=1B2024.0943.4
MM1-30BModel Scale=>14B2024.0942.8
MM1-3BModel Scale=3B2024.0941.4