List your company
ResearchFormerly WizWand

Multimodal Reasoning on MMMU (test)

64.7AccuracyGPT-4o
20.39631.89843.454.902Dec 6, 2024Feb 17, 2025May 1, 2025Jul 14, 2025Sep 25, 2025Dec 7, 2025Feb 19, 2026
Updated 4mo ago

Evaluation results

MethodLinks
GPT-4oSize=—2026.0164.7
InternVL2.5-78Bparameters=78B2024.1261.8
Qwen2.5-VL-DPParameters=7B, Post-training Strategy=Direct Preference (DP)2025.0759.4
Qwen2.5-VL-7BParameters=7B2025.0758.6
InternVL2.5-38Bparameters=38B2024.1257.6
InternVL2-Llama3-76Bparameters=76B, backbone=Llama32024.1255.1
NVLM-D-72Bparameters=72B2024.1254.6
InternVL2.5-26Bparameters=26B2024.1251.8
Full Prec2025.0850.44
Qwen2-VL-DPParameters=7B, Post-training Strategy=Direct Preference (DP)2025.0749.4
InternVL2-40Bparameters=40B2024.1249.3
InternVL2.5-8Bparameters=8B2024.1248.9
Qwen2-VL-7BParameters=7B2025.0747.8
VILA-1.5-40Bparameters=40B2024.1246.9
Qwen2-VLSize=8B2026.0146.6
InternVL2.5-4Bparameters=4B2024.1246.3
LinMU-NVSize=8B2026.0144.6
NVILASize=8B2026.0144.4
InternVL2-8Bparameters=8B2024.1244.3
InternVL2-26Bparameters=26B2024.1243.8
LLaVA-OVSize=7B2026.0142.8
InternVL2Size=8B2026.0142.6
InternVL2-4Bparameters=4B2024.1241.4
InternVL-Chat-V1.52024.1241
InternVL2.5-2Bparameters=2B2024.1238.2
LLaVA-1.5Parameters=13B2025.0736.4
InternVL2.5-1Bparameters=1B2024.1235.8
InternVL2-2Bparameters=2B2024.1234.7
VLMQ2025.0833.67
SPHINXParameters=13B2025.0732.9
InstructBLIPParameters=7B2025.0732.9
InternVL2-1Bparameters=1B2024.1232.8
GPTQ2025.0832.22
EntropyPruneRetained Tokens=320 (↓ 88.9%)2026.0230
FastVRetained Tokens=320 (↓ 88.9%)2026.0229.3
LLaVA-Next-7BRetained Tokens=2880 (100%)2026.0228
PDropRetained Tokens=320 (↓ 88.9%)2026.0227.3
CDPrunerRetained Tokens=320 (↓ 88.9%)2026.0227.3
Frequent Guess2025.0726.8
DARTRetained Tokens=320 (↓ 88.9%)2026.0226.7
DivPruneRetained Tokens=320 (↓ 88.9%)2026.0225.3
GPTAQ2025.0822.25
Random Chance2025.0722.1