List your company
ResearchFormerly WizWand

Multi-image reasoning on Muirbench (test)

68AccuracyGPT-4V
21.7233.73545.7557.765Sep 30, 2024Dec 12, 2024Feb 24, 2025May 9, 2025Jul 22, 2025Oct 4, 2025Dec 17, 2025
Updated 8mo ago

Evaluation results

MethodLinks
GPT-4VModel Scale=>14B, Note=Result is from GPT-4o as per table dagger symbol2024.0968
Qwen2.5VLSize=7B, Type=AR, Samples=>9M2025.1259.6
MM1.5-30BModel Scale=>14B2024.0958.2
MM1.5-7BModel Scale=7B2024.0949.1
LLaDA-VSize=8B, Type=Diff., Samples=16.5M2025.1248.3
Qwen2.5VLSize=3B, Type=AR, Samples=>9M2025.1247.7
DiffusionVLSize=3B, Type=Diff., Samples=738K2025.1247.2
MM1.5-3B-MoEModel Scale=3B2024.0945.6
DiffusionVLSize=7B, Type=Diff., Samples=738K2025.1244.8
Mantis-Idefics2-8BModel Scale=7B2024.0944.5
MM1.5-3BModel Scale=3B2024.0944.3
LLaVA-OVSize=7B, Type=AR, Samples=7.8M2025.1241.8
MM1.5-1B-MoEModel Scale=1B2024.0940.9
LLaVA-NeXT-Interleave-7BModel Scale=7B2024.0938.9
Phi-3-Vision-4BModel Scale=3B2024.0938
MM1-30BModel Scale=>14B2024.0936.7
MM1.5-1BModel Scale=1B2024.0934.7
Emu2-Chat-37BModel Scale=>14B2024.0933.6
MM1-1BModel Scale=1B2024.0930.7
MM1-7BModel Scale=7B2024.0930.4
MM1-3BModel Scale=3B2024.0928
Idefics2-8BModel Scale=7B2024.0926.1
LLaVAOneVision-0.5BModel Scale=1B2024.0925.5
LLaVA-v1.5-7BModel Scale=7B2024.0923.5