List your company
ResearchFormerly WizWand

Multimodal Reasoning on MMK12

81.8AccuracyVPPO-RL-7B
40.92851.53962.1572.761Dec 20, 2025Jan 19, 2026Feb 19, 2026Mar 22, 2026Apr 21, 2026May 22, 2026Jun 22, 2026
Updated 4mo ago

Evaluation results

MethodLinks
VPPO-RL-7BModel Scale=7B2026.0581.8
EASE-7BModel Scale=7B2026.0581.6
VGPO-7BModel Scale=7B2026.0581.5
PAPO_D-7BModel Scale=7B2026.0580.8
PAPOGBackbone=Qwen2.5-VL-7B2026.0672.35
GRPOBackbone=Qwen2.5-VL-7B2026.0672.26
CFPOGBackbone=Qwen2.5-VL-7B, Hyperparameters=gamma=0.02 + No Ent2026.0671.61
VL-Rethinker-7BModel Scale=7B2026.0566
MSSRModel=Qwen2.5-VL-7B, Training Protocol=Zero RL2025.1262.5
MM-Eureka-Qwen-7BTraining Protocol=Zero RL2025.1261.7
MM-Eureka-7BModel Scale=7B2026.0561.7
ThinkLite-VL-7BTraining Protocol=Zero RL2025.1257.6
RLOOModel=Qwen2.5-VL-7B, Training Protocol=Zero RL2025.1256
GRPOModel=Qwen2.5-VL-7B, Training Protocol=Zero RL2025.1255.8
OpenVLThinker-7BTraining Protocol=SFT + RL2025.1253.5
VLAA-Thinker-7BTraining Protocol=SFT + RL2025.1251.7
REINFORCE++Model=Qwen2.5-VL-7B, Training Protocol=Zero RL2025.1251.3
NoisyRollout-7BModel Scale=7B2026.0551.2
ThinkLite-VL-7BModel Scale=7B2026.0550.5
MSSRModel=Qwen2.5-VL-3B, Training Protocol=Zero RL2025.1249.2
PAPOGBackbone=Qwen3-VL-2B-Thinking2026.0648.57
Qwen2.5-VL-7BTraining Protocol=Zero RL2025.1248.1
CFPOGBackbone=Qwen3-VL-2B-Thinking2026.0648.04
GRPOBackbone=Qwen3-VL-2B-Thinking2026.0647.71
GRPOModel=Qwen2.5-VL-3B, Training Protocol=Zero RL2025.1246.1
REINFORCE++Model=Qwen2.5-VL-3B, Training Protocol=Zero RL2025.1246
RLOOModel=Qwen2.5-VL-3B, Training Protocol=Zero RL2025.1245.5
R1-Onevision-7BTraining Protocol=SFT + RL2025.1243.5
Qwen2.5-VL-3BTraining Protocol=Zero RL2025.1242.5