List your company
ResearchFormerly WizWand

Multi-discipline Multimodal Understanding on MMMU Pro

67.3AccuracyGPT-5-Mini
17.69230.57143.4556.329Aug 25, 2025Sep 26, 2025Oct 28, 2025Nov 29, 2025Dec 31, 2025Feb 1, 2026Mar 6, 2026
Updated 5mo ago

Evaluation results

MethodLinks
GPT-5-Mini2025.1267.3——
InternVL3.5-38B2025.1266——
Qwen3-VL-32B2025.1265.3——
GLM-4.5V2025.1265.2——
InternVL3-78B2025.1262.3——
Claude-4-Sonnet2025.1261.6——
Qwen3-VL-8B-MastersSize=8B2025.1261.4——
MastersBase Model=Qwen3-VL-8B2025.1261.4——
Qwen2.5-VL-72B2025.1261.2——
Qwen2.5-VL-7B-MastersSize=7B2025.1260.6——
MastersBase Model=Qwen2.5-VL-7B2025.1260.6——
InternVL3-8B-MastersSize=8B2025.1258.8——
MastersBase Model=InternVL3-8B2025.1258.8——
InternVL3.5-8B-MastersSize=8B2025.1258.1——
MastersBase Model=InternVL3.5-8B2025.1258.1——
InternVL3.5-8BSize=8B2025.1257.2——
Qwen3-VL-4B-MastersSize=4B2025.1257——
Claude-3.7-Sonnet2025.1256.5——
Qwen3-VL-8BSize=8B2025.1255.9——
Qwen3-VLModel Size=8B2026.0355.9——
Active-ZeroBackbone=Qwen2.5-VL-7B-Instruct2026.0254.62——
Gemini-2.0-Flash2025.1254.4——
Qwen3-VL-4BSize=4B2025.1253.2——
InternVL3.5-4B-MastersSize=4B2025.1252.7——
Base ModelBackbone=Qwen2.5-VL-7B-Instruct2026.0252.61——
InternVL3.5-2B-MastersSize=2B2025.1252.4——
VisionZero-RealWorldBackbone=Qwen2.5-VL-7B-Instruct2026.0252.29——
VisionZero-CLEVRBackbone=Qwen2.5-VL-7B-Instruct2026.0252.17——
VisPlayBackbone=Qwen2.5-VL-7B-Instruct2026.0251.98——
GPT-4o2025.1251.9——
VisionZero-ChartBackbone=Qwen2.5-VL-7B-Instruct2026.0251.67——
EvoLMMBackbone=Qwen2.5-VL-7B-Instruct2026.0251.6——
Claude-3.5-Sonnet2025.1251.5——
InternVL3.5-4BSize=4B2025.1251.4——
Qwen2.5-VL-3B-MastersSize=3B2025.1251——
Gemini-1.5-Pro2025.1246.9——
Active-ZeroBackbone=Qwen2.5-VL-3B-Instruct2026.0246.7——
InternVL3.5-2BSize=2B2025.1246.3——
MiMo-VL-8BSize=8B2025.1246.2——
VisPlayBackbone=Qwen2.5-VL-3B-Instruct2026.0242.43——
Base ModelBackbone=Qwen2.5-VL-3B-Instruct2026.0241.99——
InternVL3-8BSize=8B2025.1241.3——
Penguin-VLModel Size=8B2026.0340.2——
InternVL3-2B-MastersSize=2B2025.1240.1——
VideoAuto-R1Backbone=Qwen2.5-VL-7B2026.0139.8——
InternVL-3.5Model Size=8B2026.0339.7——
Qwen3-VL-2B-MastersSize=2B2025.1239.5——
Phi-4-Multimodal-5.6BSize=5.6B2025.1238.5——
Qwen2.5-VL-7BSize=7B2025.1238.3——
LLaVA-OneVision-1.5-8BSize=8B2025.1237.4——
Qwen3-VL-2BSize=2B2025.1236.5——
OpenAI GPT-5 nanoModel Size=nano2026.0336.5——
Qwen2.5-VL-7BBackbone=Qwen2.5-VL-7B2026.0136.1——
LLaVA-OneVision-1.5-4BSize=4B2025.1235.3——
Qwen2.5-VL-3BSize=3B2025.1231.6——
LLaVA-OneVision-72B2025.1231——
BaseTraining Stage=Base, RL Alignment (GRPO)=false2025.0829.59——
BaseTraining Stage=Base, RL Alignment (GRPO)=true2025.0828.61——
PSFTTraining Stage=PSFT, RL Alignment (GRPO)=false2025.0827.63——
MiniCPM-V2.6-8BSize=8B2025.1227.2——
PSFTTraining Stage=PSFT, RL Alignment (GRPO)=true2025.0826.07——
InternVL3-2BSize=2B2025.1224.9——
LLaVA-OneVision-7BSize=7B2025.1224.1——
SFTTraining Stage=SFT, RL Alignment (GRPO)=false2025.0823.58——
Phi-3.5-Vision-4BSize=4B2025.1219.7——
SFTTraining Stage=SFT, RL Alignment (GRPO)=true2025.0819.6——
AVAR-ThinkerModel Category=Our model2026.03—42.9—
Cambrian-1LLM Backbone=LLaMA3-8B, Open Data=true2025.12—14.7—
Claude 3.5 SonnetLLM Backbone=-, Open Data=false2025.12—48—
Claude-3.7-SonnetModel Category=Closed-Source2026.03—50.1—
COMPACTRecipe=COMPACT2025.04—20.2311.91
Dream-VLLLM Backbone=Dream 7B, Open Data=true, Input Mode=Single Image2025.12—25.8—
Dream-VLLLM Backbone=Dream 7B, Open Data=true2025.12—26—
Gemini-1.5 ProLLM Backbone=-, Open Data=false2025.12—44.4—
GPT-4oLLM Backbone=-, Open Data=false2025.12—49.7—
GPT-4oModel Category=Closed-Source2026.03—54.5—
InternVL2.5-8BModel Category=Open-Source General Models2026.03—38.2—
LLaDA-VLLM Backbone=LLaDA 8B, Open Data=true2025.12—18.6—
Llama-3.2-11B-Vision-InstructModel Category=Open-Source General Models2026.03—33—
LLaVA-665KRecipe=LLaVA-665K2025.04—20.1211.97
Llava-CoT-11BLLM Backbone=LLaMA3.2-11B, Open Data=true2025.12—18.5—
LLaVA-OneVision-7BModel Category=Open-Source General Models2026.03—35.5—
LLaVA-OVLLM Backbone=Qwen2-7B, Open Data=true, Input Mode=Single Image2025.12—16.8—
LLaVA-OVLLM Backbone=Qwen2-7B, Open Data=true2025.12—18.7—
MAmmoTH-VLLLM Backbone=Qwen2.5-7B, Open Data=true, Input Mode=Single Image2025.12—26—
MAmmoTH-VLLLM Backbone=Qwen2.5-7B, Open Data=true2025.12—25.3—
MiMo-VL-RLLLM Backbone=MiMo-7B, Open Data=false2025.12—40.3—
MM-Eureka-7BModel Category=Multimodal Reasoning Models2026.03—42.4—
Molmo-8B-DLLM Backbone=Qwen2 7B, Open Data=true2025.12—18.9—
Mulberry-7BModel Category=Multimodal Reasoning Models2026.03—34.8—
OpenVLThinkerModel Category=Multimodal Reasoning Models2026.03—42.9—
Qwen2.5-VL-7BModel Category=Open-Source General Models2026.03—38.3—
R1-OneVisionModel Category=Multimodal Reasoning Models2026.03—32.2—
RandomRecipe=Random2025.04—18.1511.44
ThinkLite-VLModel Category=Multimodal Reasoning Models2026.03—40—
Vision-R1Model Category=Multimodal Reasoning Models2026.03—39.6—
Vision-SR1Model Category=Multimodal Reasoning Models2026.03—43.8—
VLAA-Thinker-7BModel Category=Multimodal Reasoning Models2026.03—40.9—