List your company
ResearchFormerly WizWand

Logical Reasoning on LogicVista (pass@1/avg@8)

64.9Avg Pass@8ADHint
21.01232.40643.855.194Jul 8, 2025Aug 20, 2025Oct 2, 2025Nov 14, 2025Dec 27, 2025Feb 8, 2026Mar 24, 2026
Updated 6mo ago

Evaluation results

MethodLinks
ADHintBase Model=Qwen3-VL-8B-Instruct2025.1264.956.3
GRPOBase Model=Qwen3-VL-8B-Instruct2025.126349.4
Qwen2.5-VL-32B + VPPOModel Scale=32B, Algorithm=VPPO2025.1059.2—
Qwen2.5-VL-32B + DAPOModel Scale=32B, Algorithm=DAPO2025.1058.9—
Qwen2.5-VL-32B + GRPOModel Scale=32B, Algorithm=GRPO2025.1058.3—
MM-Eureka-32BModel Scale=32B, Prompting Source=Official author-provided2025.1056.8—
NoisyRollout-32BModel Scale=32B, Prompting Source=Official author-provided, Note=Trained using the training set of Geo3k2025.1056.2—
Qwen2.5-VL-32BModel Scale=32B2025.1052.8—
HintGRPOBase Model=Qwen3-VL-8B-Instruct2025.1251.943.1
ADHintBase MLLM=Qwen2.5-VL-7B2025.1249.448.7
Qwen2.5-VL-7B + VPPOModel Scale=7B, Algorithm=VPPO2025.1047.9—
SFT -> RLBase MLLM=Qwen2.5-VL-7B2025.1247.828.8
NoisyRollout-7BModel Scale=7B, Training Strategy=Pure RL, Prompting Source=Official author-provided, Note=Trained using the training set of Geo3k2025.1047.3—
Qwen2.5-VL-7B + DAPOModel Scale=7B, Algorithm=DAPO2025.1046.8—
PAPO-D-7BModel Scale=7B, Training Strategy=Pure RL2025.1046.7—
PAPO_DBackbone=Qwen2.5-VL, Model Size=7B2025.0746.7—
GRPOBase MLLM=Qwen2.5-VL-7B2025.1246.444.9
MM-Eureka-7BModel Scale=7B, Training Strategy=Pure RL, Prompting Source=Official author-provided2025.1046.3—
VL-Rethinker-7BModel Scale=7B, Training Strategy=Pure RL, Prompting Source=Official author-provided2025.1046.3—
PAPO_GBackbone=Qwen2.5-VL, Model Size=7B2025.0746.07—
SFTBase MLLM=Qwen2.5-VL-7B2025.124623.7
GRPOBackbone=Qwen2.5-VL, Model Size=7B2025.0745.62—
R1-ShareVL-7BModel Scale=7B, Training Strategy=Pure RL, Prompting Source=Official author-provided2025.1045.6—
Qwen2.5-VL-7B + GRPOModel Scale=7B, Algorithm=GRPO2025.1045.6—
Qwen2.5-VL-7BModel Scale=7B2025.1042.4—
PAPO_DBackbone=Qwen2.5-VL, Model Size=3B2025.0741.67—
ADHintBase MLLM=Qwen2.5-VL-3B2025.1241.442.6
DAPOBackbone=Qwen2.5-VL, Model Size=3B2025.0740.69—
GHPOBase MLLM=Qwen2.5-VL-7B2025.1240.542.4
SFT -> RLBase MLLM=Qwen2.5-VL-3B2025.1239.918.1
PEPO_GBackbone=Qwen2.5-VL-3B-Instruct, Training Data=ViRL39K2026.0339.85—
ThinkLite-7BModel Scale=7B, Training Strategy=Pure RL, Prompting Source=Official author-provided2025.1039.4—
LUFFYBase MLLM=Qwen2.5-VL-7B2025.1239.240.2
GRPOBase MLLM=Qwen2.5-VL-3B2025.1238.943.8
PAPO_GBackbone=Qwen2.5-VL-3B-Instruct, Training Data=ViRL39K2026.0338.67—
PAPO_GBackbone=Qwen2.5-VL, Model Size=3B2025.0738.67—
LUFFYBase MLLM=Qwen2.5-VL-3B2025.1238.335.4
GRPOBackbone=Qwen2.5-VL-3B-Instruct, Training Data=ViRL39K2026.0338.14—
GRPOBackbone=Qwen2.5-VL, Model Size=3B2025.0738.14—
SFTBase MLLM=Qwen2.5-VL-3B2025.1237.818.1
DAPOBackbone=Qwen2.5-VL, Model Size=7B2025.0737.05—
HintGRPOBase MLLM=Qwen2.5-VL-7B2025.1234.736.4
GHPOBase MLLM=Qwen2.5-VL-3B2025.1233.935.3
HintGRPOBase MLLM=Qwen2.5-VL-3B2025.1233.735.7
PAPO_GBackbone=Qwen3-VL (thinking), Model Size=2B2025.0732.83—
GRPOBackbone=Qwen3-VL (thinking), Model Size=2B2025.0729.84—
StepHintBase MLLM=Qwen2.5-VL-3B2025.1227.812.5
StepHintBase MLLM=Qwen2.5-VL-7B2025.1222.710.5