List your company
ResearchFormerly WizWand

Multimodal Math Reasoning on MathVision

86AccuracyQwen3.5-27B
28.373643.334358.29573.2557Feb 4, 2025Apr 26, 2025Jul 17, 2025Oct 6, 2025Dec 27, 2025Mar 18, 2026Jun 8, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Qwen3.5-27BMode=REASONING, Architecture=Dense, # Total Params=27B, # Activated Params=27B2026.0486
EXAONE 4.5 33BMode=REASONING, Architecture=Dense, # Total Params=33B, # Activated Params=33B2026.0475.2
Qwen3-VL-235B-A22BMode=Thinking, Architecture=MoE, # Total Params=236B, # Activated Params=23B2026.0474.6
GPT-5 miniMode=REASONING: HIGH, Architecture=-, # Total Params=-, # Activated Params=-2026.0471.9
GPT-5 highInput Modality=Multimodal, LLM-as-a-Judge=GPT-4o2025.1171.6
Qwen3-VL-32BMode=Thinking, Architecture=Dense, # Total Params=33B, # Activated Params=33B2026.0470.2
Seed-1.5-thinkingAccess Type=Closed-Source2025.1268.7
Gemini 2.5 ProInput Modality=Multimodal, LLM-as-a-Judge=GPT-4o2025.1163.3
o12026.0360.3
o1Model Category=Closed-Source Models2026.0360.3
Claude Sonnet 4.5Input Modality=Multimodal, LLM-as-a-Judge=GPT-4o2025.1158.7
InternVL3.5-8BModel Scale=8B, Optimization Method=N/A2026.0656.8
Gemini-2.5-Pro-ThinkingAccess Type=Closed-Source2025.1255.3
Claude-4-SonnetAccess Type=Closed-Source2025.1254.6
Qwen2.5VL-72B x Qwen3-32BParam (B)=1042025.0952.6
GPT-4.1Access Type=Closed-Source2025.1251.8
Qwen3-VLModel Size=4B, Training Stage=instruct2026.0551.6
InternVL3.5-4B (+GNDPO)Model Scale=4B, Optimization Method=GNDPO2026.0651.6
Vision-R1Backbone=Vision-R1 (7B)2026.0551.2
GLM-4v-Plus-20250111Param (B)=-2025.0951.1
InternVL3.5-4B (+OPD)Model Scale=4B, Optimization Method=OPD2026.0650.5
Qwen3-VL-SegModel Size=4B, Training Stage=S-22026.0550.4
InternVL3.5-4B (+GSPO)Model Scale=4B, Optimization Method=GSPO2026.0648.7
Gemini-2.0-ProModel Category=Proprietary Models2025.0648.1
Qwen3-VLModel Size=4B, Training Stage=S-12026.0547.9
Sora-2 AudioInput Modality=Audio, LLM-as-a-Judge=GPT-4o2025.1146.7
Qwen2.5VL-7B x Qwen3-32BParam (B)=392025.0946.6
GPT-4.1-20250414Param (B)=-2025.0945.1
Sora-2 Last FrameInput Modality=Last Frame, LLM-as-a-Judge=GPT-4o2025.1144.9
PDCRBackbone=Qwen2.5-VL-7B2026.0544.8
Revisual-R1Access Type=Open-Source, Model Scale=7B2025.1244.7
PACRBackbone=Qwen2.5-VL-7B2026.0544.7
OursAccess Type=Open-Source, Model Scale=7B2025.1244.2
DAPOBackbone=Qwen2.5-VL-7B2026.0544.2
ChatGPT-4o-latestParam (B)=-2025.0943.8
GPT-4oModel Category=Proprietary Models2025.0643.8
Qwen2.5VL-7B x Qwen3-4BParam (B)=112025.0943.2
Claude3.7-SonnetParam (B)=-2025.0941.9
Claude-3.7-SonnetModel Category=Proprietary Models2025.0641.9
Claude3.7-Sonnet2026.0341.3
Gemini2-flash2026.0341.3
Claude3.7-SonnetModel Category=Closed-Source Models2026.0341.3
Gemini-2-flashModel Category=Closed-Source Models2026.0341.3
Claude-3.7-SonnetModel Category=Close-source Models2026.0441.3
GRPOBackbone=Qwen2.5-VL-7B2026.0541.3
InternVL3.5-4B (Base (Instruct))Model Scale=4B, Optimization Method=Base (Instruct)2026.0640.5
Zero-shot InferenceBackbone=Qwen2.5-VL-7B, Mode=Zero-shot2026.0540.3
Qwen-2.5-VL-32B2026.0340.1
Qwen-2.5-VL-32BModel Category=Open-Source General Models2026.0340.1
PDCRBackbone=Qwen2.5-VL-3B2026.0540.1
MM-Eureka-32B-R-TAPModel Category=Open-Source Reasoning Models2026.0339.9
Gemma-3-27BParameters=27B2025.1239.8
GRPOBackbone=Qwen2.5-VL-3B2026.0539.4
Qwen2.5-VL-72BParam (B)=722025.0939.3
Qwen2.5-VL-72BModel Category=Open-source Models (>70B)2025.0639.3
DAPOBackbone=Qwen2.5-VL-3B2026.0539.1
PACRBackbone=Qwen2.5-VL-3B2026.0539
InternVL3-78BParam (B)=782025.0938.8
Zero-shot InferenceBackbone=Qwen2.5-VL-3B, Mode=Zero-shot2026.0538.6
InternVL3.5-2B (+GNDPO)Model Scale=2B, Optimization Method=GNDPO2026.0638.4
Qwen2.5-VL-72BModel Category=Larger MLLMs without Reasoning2025.1238.1
Qwen-2.5-VL-72B2026.0338.1
Qwen-2.5-VL-72BModel Category=Open-Source General Models2026.0338.1
InternVL3.5-2B (+OPD)Model Scale=2B, Optimization Method=OPD2026.0637.9
GPT-4oModel Category=Larger MLLMs without Reasoning2025.1236.5
Visionary-R1Backbone=Visionary-R1 (3B)2026.0536.5
InternVL2.5-78B-MPOModel Category=Open-source Models (>70B)2025.0636.2
RISEBackbone=Qwen3-VL-8B-Instruct, Self-evolving steps=602026.0536.1
QVQ-72B-Preview2026.0335.9
QVQ-72B-PreviewModel Category=Open-Source Reasoning Models2026.0335.9
InternVL3.5-2B (+GSPO)Model Scale=2B, Optimization Method=GSPO2026.0635.9
Perception-R1Backbone=Perception-R1 (7B)2026.0535.7
Qwen2.5-VL-7B-Instruct + R-TAPModel Size=7B2026.0335.3
QVQ-72B-PreviewParam (B)=722025.0934.9
Qwen2.5-VL-7B + SRPOBackbone=Qwen2.5-VL-7B2026.0534.54
MM-Eureka-32BModel Category=Open-Source Reasoning Models2026.0334.4
ThinkLite-VLAccess Type=Open-Source, Model Scale=7B2025.1232.9
RISEBackbone=Qwen3-VL-8B-Instruct, Self-evolving steps=402026.0532.8
PeBR-R1Access Type=Open-Source, Model Scale=7B2025.1232.7
AStar (Qwen2.5-7B)OS Only=✓, Training-Free=✓, Prior Data=0.5K, Pre. Time=50 mins2025.0232.7
VL-RethinkerAccess Type=Open-Source, Model Scale=7B2025.1232.3
InternVL2.5-38B-MPO2026.0332.3
InternVL2.5-38B-MPOModel Category=Open-Source Reasoning Models2026.0332.3
InternVL2.5-VL-78B2026.0332.2
InternVL2.5-VL-78BModel Category=Open-Source General Models2026.0332.2
InternVL2.5-78BModel Category=Open-source Models (>70B)2025.0632.2
DIVA-GRPO-7B2026.0332.1
VAPO-ThinkerAccess Type=Open-Source, Model Scale=7B2025.1231.9
InternVL2.5-VL-38B2026.0331.8
InternVL2.5-VL-38BModel Category=Open-Source General Models2026.0331.8
MM-Eureka-7B-R-TAPModel Category=Open-Source Reasoning Models2026.0331.7
ToR-DAPOData Size=39K (ViRL-39K), Base Model=Qwen-2.5-VL-7B2026.0331.6
Vision-G1Access Type=Open-Source, Model Scale=7B2025.1231.3
RISEBackbone=Qwen3-VL-8B-Instruct, Self-evolving steps=202026.0531.28
GPT-4oParam (B)=-2025.0931.2
PRCO-7BBackbone=Qwen2.5-VL-7B2026.0530.92
Qwen2.5-VL-7B + GRPOBackbone=Qwen2.5-VL-7B2026.0530.92
SelfJudgeTraining Data=Geo3K, Backbone=Qwen2.5-VL-7B2026.0330.9
R1-OnevisionParam (B)=72025.0930.6
VL-Rethinker-7BBackbone=Qwen2.5-VL-7B2026.0530.59