List your company
ResearchFormerly WizWand

Image Understanding on TextVQA

725AccuracyMiniCPM-LLaMA3-V 2.5
18.112201.631385.15568.669Oct 27, 2025Dec 1, 2025Jan 5, 2026Feb 9, 2026Mar 16, 2026Apr 20, 2026May 25, 2026
Updated 5mo ago

Evaluation results

MethodLinks
MiniCPM-LLaMA3-V 2.5Size=8B2026.03725
Top-k RoutingExpert skipping ratio (ρ)=0, Number of experts (k)=8, Backbone=InternVL-3.5-30B-A3B-HF2025.1185.76
MC-MoEExpert skipping ratio (ρ)=0.6, Number of experts (k)=3, Backbone=InternVL-3.5-30B-A3B-HF2025.1184.36
MoDESExpert skipping ratio (ρ)=0.6, Number of experts (k)=3, Backbone=InternVL-3.5-30B-A3B-HF2025.1184.27
DiEPExpert skipping ratio (ρ)=0.6, Number of experts (k)=3, Backbone=InternVL-3.5-30B-A3B-HF2025.1183.68
NAEEExpert skipping ratio (ρ)=0.6, Number of experts (k)=3, Backbone=InternVL-3.5-30B-A3B-HF2025.1182.98
Top-k RoutingExpert skipping ratio (ρ)=0.6, Number of experts (k)=3, Backbone=InternVL-3.5-30B-A3B-HF2025.1182.16
MoDESExpert skipping ratio (ρ)=0.73, Number of experts (k)=2, Backbone=InternVL-3.5-30B-A3B-HF2025.1182.13
Qwen2.5-VLSize=7B2026.0381.2
Insight-V++Size=7B, Base Model=Qwen2.5-VL, Self-Evolving=true2026.0380.6
MoDESExpert skipping ratio (ρ)=0.85, Number of experts (k)=1, Backbone=InternVL-3.5-30B-A3B-HF2025.1180.58
Qwen2.5-VL + Multi-Agent (RL)Size=7B, Multi-Agent (RL)=true2026.0380.4
MiniCPM-V-2.6Size=7B2026.0378.3
MC-MoEExpert skipping ratio (ρ)=0.73, Number of experts (k)=2, Backbone=InternVL-3.5-30B-A3B-HF2025.1177.41
Our Base Model + Multi-AgentSize=7B, Multi-Agent=true2026.0377
DiEPExpert skipping ratio (ρ)=0.73, Number of experts (k)=2, Backbone=InternVL-3.5-30B-A3B-HF2025.1176.84
Insight-VSize=7B, Base Model=Our Base Model, Iterative DPO=true2026.0376.8
Our Base ModelSize=7B2026.0375.4
NAEEExpert skipping ratio (ρ)=0.73, Number of experts (k)=2, Backbone=InternVL-3.5-30B-A3B-HF2025.1175.37
Ovis1.5-LLaMA3Size=8B2026.0374
Idefics3-LLaMA3Size=8B2026.0373.2
Cambrian-1Size=8B2026.0372.6
Insight-V-LLaVASize=8B, Base Model=LLaVA-NeXT-LLaMA3, Iterative DPO=true2026.0370.5
MC-MoEExpert skipping ratio (ρ)=0.85, Number of experts (k)=1, Backbone=InternVL-3.5-30B-A3B-HF2025.1170.41
InternLM-XComposer2Size=7B2026.0369.7
DiEPExpert skipping ratio (ρ)=0.85, Number of experts (k)=1, Backbone=InternVL-3.5-30B-A3B-HF2025.1169.37
LLaVA-NeXT-LLaMA3 + Multi-AgentSize=8B, Multi-Agent=true2026.0368.9
NAEEExpert skipping ratio (ρ)=0.85, Number of experts (k)=1, Backbone=InternVL-3.5-30B-A3B-HF2025.1166.24
LLaVA-NeXT-LLaMA3Size=8B2026.0365.3
Top-k RoutingExpert skipping ratio (ρ)=0.73, Number of experts (k)=2, Backbone=InternVL-3.5-30B-A3B-HF2025.1164.51
Bunny-LLaMA3Size=8B2026.0362.1
POINTSSize=7B2026.0360
Top-k RoutingExpert skipping ratio (ρ)=0.85, Number of experts (k)=1, Backbone=InternVL-3.5-30B-A3B-HF2025.1158.49
Full-FinetuneDataset=Vision-Flan-186K, #Data=186k2026.0550.4
MHRoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=MHRoPE2025.1048.6
MRoPE-IBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=MRoPE-I2025.1048.27
MRoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=MRoPE2025.1048.25
CircleRoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=CircleRoPE2025.1047.27
HoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=HoPE2025.1047.16
VideoRoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=VideoRoPE2025.1047.1
Vanilla RoPEBackbone Model=Qwen3-VL-4B-Instruct, Positional Encoding Variant=Vanilla RoPE2025.1046.65
MAGICDataset=Vision-Flan-186K, #Data=∼37k2026.0545.7
RandomDataset=Vision-Flan-186K, #Data=∼37k2026.0545.3