List your company
ResearchFormerly WizWand

Mathematical Reasoning on MathVista (Acc@1/Acc@4)

74.2AccuracyQwen2.5-VL-7B + DUPL
46.95254.02661.168.174Oct 1, 2025Nov 14, 2025Dec 28, 2025Feb 11, 2026Mar 27, 2026May 10, 2026Jun 24, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Qwen2.5-VL-7B + DUPLBase Model=Qwen2.5-VL-7B, RL Algorithm=DUPL2025.1074.2——
PAPO-7BBase Model=PAPO-7B2025.1073.7——
Qwen2.5-VL-7B + NoisyRolloutBase Model=Qwen2.5-VL-7B, RL Algorithm=NoisyRollout2025.1072.6——
MM-Eureka-Qwen-7BBase Model=MM-Eureka-Qwen-7B2025.1071.2——
Qwen2.5-VL-7B + GRPOBase Model=Qwen2.5-VL-7B, RL Algorithm=GRPO2025.1070.6——
OpenVLThinker-7BBase Model=OpenVLThinker-7B2025.1070——
Qwen2.5-VL-3B + DUPLBase Model=Qwen2.5-VL-3B, RL Algorithm=DUPL2025.1069.4——
VLAA-Thinker-7BBase Model=VLAA-Thinker-7B2025.1068.3——
Qwen2.5-VL-3B + NoisyRolloutBase Model=Qwen2.5-VL-3B, RL Algorithm=NoisyRollout2025.1067.7——
Qwen2.5-VL-7BBase Model=Qwen2.5-VL-7B2025.1067.2——
CFPOGBackbone=Qwen3-VL-2B-Thinking2026.0666.96——
CoresetBackbone=Qwen2.5-VL-7B, Training=finetuned2025.1066.9——
RECAPBackbone=Qwen2.5-VL-7B, Training=finetuned2025.1066.7——
Qwen2.5-VL-3B + GRPOBase Model=Qwen2.5-VL-3B, RL Algorithm=GRPO2025.1066.4——
UniformBackbone=Qwen2.5-VL-7B, Training=finetuned2025.1065.6——
Reasoning-onlyBackbone=Qwen2.5-VL-7B, Training=finetuned2025.1065.5——
R1-Onevision-7BBase Model=R1-Onevision-7B2025.1063.9——
VLAA-Thinker-7B2025.1063.9——
LwFBackbone=Qwen2.5-VL-7B, Training=finetuned2025.1063.9——
PropMixBackbone=Qwen2.5-VL-7B, Training=finetuned2025.1063.4——
Vision-R1-7B2025.1061.8——
Qwen2.5-VL-7Bvariant=Base model2025.1061.7——
OpenVLThinker-7B2025.1059.1——
Qwen2.5-VL-3BBase Model=Qwen2.5-VL-3B2025.1058.4——
SFT + VIGILBase Model=Qwen2.5-VL-72B2026.0656.6——
PAPOGBackbone=Qwen3-VL-2B-Thinking2026.0656.08——
SFT + DA-DPOBase Model=Qwen2.5-VL-72B2026.0655.4——
SFT + SimPOBase Model=Qwen2.5-VL-72B2026.0655.2——
SFT onlyBase Model=Qwen2.5-VL-72B2026.0654.5——
SFT + DPOBase Model=Qwen2.5-VL-72B2026.0654.1——
SFT + HA-DPOBase Model=Qwen2.5-VL-72B2026.0653.8——
GRPOBackbone=Qwen3-VL-2B-Thinking2026.0653.58——
SFT + VIGILBase Model=Qwen2.5-VL-7B2026.0649.5——
SFT + SimPOBase Model=Qwen2.5-VL-7B2026.0649.1——
SFT + DA-DPOBase Model=Qwen2.5-VL-7B2026.0648.8——
SFT onlyBase Model=Qwen2.5-VL-7B2026.0648.2——
SFT + DPOBase Model=Qwen2.5-VL-7B2026.0648——
Gemini-2.5-Pro*Model Category=Closed-Source Models2026.04—80.985.2
GPT-5-Thinking*Model Category=Closed-Source Models2026.04—81.986.1
InternVL2.5-8B*Model Category=Open-Source Models2026.04—68.272.8
MUPO-Thinker-3BModel Category=Our Models2026.04—64.372.5
MUPO-Thinker-7BModel Category=Our Models2026.04—77.982.4
Qwen2.5-VL-7BModel Category=Open-Source Models2026.04—62.378.5
R1-OneVision-7BModel Category=Open-Source Models2026.04—64.169.4
Vision-R1-7BModel Category=Open-Source Models2026.04—73.575.6
VLAA-Thinker-7BModel Category=Open-Source Models2026.04—6870.8