List your company
ResearchFormerly WizWand

Question Answering on ScienceQA

98.8AccuracyOneVision-8B
60.73670.61880.590.382Nov 11, 2023Apr 14, 2024Sep 17, 2024Feb 20, 2025Jul 25, 2025Dec 28, 2025Jun 2, 2026
Updated 4mo ago

Evaluation results

MethodLinks
OneVision-8BModel=OneVision-8B2026.0698.8
OneVision-4BModel=OneVision-4B2026.0698.5
InternVL2-8BModel=InternVL2-8B2026.0697.2
InternVL2-4BModel=InternVL2, Parameter Count=4B, Result Source=OpenVLM Leaderboard (OpenCompass, 2025)2026.0496.3
LLaVA-OneVision-7BModel Scale=7B2024.0896
UDSModel=Qwen-2.5-7B2025.1095.19
LLaVA-OneVision-7BModel=LLaVA-OneVision, Parameter Count=7B2026.0494.71
PESC (block-wise)Backbone=Mistral-7B, Evaluation Protocol=domain-specific fine-tuning2026.0394.33
LoRABackbone=Mistral-7B, Evaluation Protocol=domain-specific fine-tuning2026.0394.15
FisherSFTModel=Qwen-2.5-7B2025.1094.02
AdaFuseBackbone=Mistral-7B, Evaluation Protocol=domain-specific fine-tuning2026.0393.82
MoRAL (layer-wise)Backbone=Mistral-7B, Evaluation Protocol=domain-specific fine-tuning2026.0393.79
Qwen3-VL-4BModel=Qwen3-VL, Parameter Count=4B2026.0493.42
InfiniteVL-4BModel=InfiniteVL, Parameter Count=4B, Result Source=InfiniteVL (Tao et al., 2025) paper2026.0493.4
Llama-90BModel=Llama-90B2026.0691.4
Single Task SoTA2023.1290.9
LLaVA-OneVision-72BModel Scale=72B2024.0890.3
Pragma-VLBackbone=Qwen2.5-VL-7B2026.0289.06
SFTBackbone=Qwen2.5-VL-7B2026.0289.01
DPOBackbone=Qwen2.5-VL-7B2026.0288.86
UIO-2XLModel Size=3.2B2023.1288.8
UIO-2XXLModel Size=6.8B2023.1288.7
Qwen2.5-VL-7BBackbone=Qwen2.5-VL-7B2026.0288.48
MM-RLHFBackbone=Qwen2.5-VL-7B2026.0287.45
SmolVLM2-2.2BModel=SmolVLM2, Parameter Count=2.2B2026.0486.6
SPA-VLBackbone=Qwen2.5-VL-7B2026.0286.32
Beavertails-V_helpBackbone=Qwen2.5-VL-7B2026.0286.06
Beavertails-V_harmBackbone=Qwen2.5-VL-7B2026.0285.93
SliMELLM=Llama3-8B2024.1184.2
HalluVL-DPOBackbone=Qwen2-VL-7B2026.0482.2
UIO-2LModel Size=1.1B2023.1281.6
SpB2.0-VL-5BModel=SpB2.0-VL, Parameter Count=5B2026.0481.55
OneVision-4B-LNModel=OneVision-4B-LN2026.0681.5
HalluVL-DPOwBackbone=Qwen2-VL-7B2026.0481.36
VCDBackbone=Qwen2-VL-7B2026.0481.21
Qwen2-VL-7BBackbone=Qwen2-VL-7B2026.0481.06
Cambrian-8BModel=Cambrian-8B2026.0680.3
Qwen2.5-VL-3BModel=Qwen2.5-VL, Parameter Count=3B2026.0479.54
InstructBLIPModel Size=8.2B2023.1279.5
FocusLLaVALLM=Llama3-8B2024.1179
LLaVA-NeXTLLM=Llama3-8B, implementation=ours2024.1177.3
GPT-4VVersion=V-Preview2024.0875.7
HiREDLLM=Vicuna-13B2024.1173.2
SeqRound Learned=R4, Task Type=New Task2026.0173
OSWRound Learned=R4, Task Type=New Task2026.0173
Naive SelectionParam. Scale=Extra Large2026.0372.9
AblationRound Learned=R4, Task Type=New Task2026.0172
SCoOPParam. Scale=Extra Large2026.0371.85
LLaVa-1.5Model Size=13B, Zero-shot=true2023.1271.6
Majority VotingParam. Scale=Medium2026.0371.53
SCoOPParam. Scale=Medium2026.0371.32
LLaVA-13BModel=LLaVA-13B2026.0671.2
Majority VotingParam. Scale=Extra Large2026.0371.11
Majority VotingParam. Scale=Large2026.0370.69
LLaVA-NeXTLLM=Vicuna-7B2024.1170.1
SCoOPParam. Scale=Large2026.0370.06
Avg. of Single ModelsParam. Scale=Extra Large2026.0369.75
Naive SelectionParam. Scale=Medium2026.0369.43
MonkeyParameters=9.8B, Resolution=896x8962023.1169.4
Cobra-3BModel=Cobra, Parameter Count=3B, Result Source=InfiniteVL (Tao et al., 2025) paper2026.0469.3
Avg. of Single ModelsParam. Scale=Large2026.0369.16
TrimLLM=Vicuna-7B2024.1169.1
Naive SelectionParam. Scale=Large2026.0368.91
Pragma-VLBackbone=Llava-1.5-7B2026.0268.75
mPLUG-Owl2Model Size=8.2B, Zero-shot=true2023.1268.7
mPLUG-Owl22023.1168.7
LLaVA-Prumerge+LLM=Vicuna-7B2024.1168.3
Qwen-VL-Chat2023.1168.2
LLaVA-7BModel=LLaVA-7B2026.0667.7
Majority VotingParam. Scale=Small2026.0367.44
SCoOPParam. Scale=Small2026.0367.33
LLaVA-OneVision-0.5BModel Scale=0.5B2024.0867.2
Qwen-VLModel Size=9.6B, Zero-shot=true2023.1267.1
Qwen-VLBackbone=Qwen-7B2023.1167.1
ReplayRound Learned=R4, Task Type=New Task2026.0167
LLaVa-1.5Model Size=7.2B, Zero-shot=true2023.1266.8
LLaVA-1.5LLM Type=Vicuna-7B2024.0366.8
LLaVA1.5Backbone=Vicuna-7B2023.1166.8
SFTBackbone=Llava-1.5-7B2026.0266.79
Naive SelectionParam. Scale=Small2026.0366.49
DPOBackbone=Llava-1.5-7B2026.0266.25
Avg. of Single ModelsParam. Scale=Medium2026.0366.05
MM-RLHFBackbone=Llava-1.5-7B2026.0266.01
Llava-1.5-7BBackbone=Llava-1.5-7B2026.0265.96
MaTVLM-3BModel=MaTVLM, Parameter Count=3B, Result Source=InfiniteVL (Tao et al., 2025) paper2026.0465.9
LLaVA-1.5-7BBackbone=LLaVA-1.5-7B2026.0465.54
SPA-VLBackbone=Llava-1.5-7B2026.0265.52
Beavertails-V_helpBackbone=Llava-1.5-7B2026.0265.23
Beavertails-V_harmBackbone=Llava-1.5-7B2026.0264.94
PSALMLLM Type=Phi-1.5 (1.3B)2024.0364.9
AntidoteBackbone=LLaVA-1.5-7B, Hyperparameter Configuration=HP-orig2026.0464.55
AntidoteBackbone=LLaVA-1.5-7B, Hyperparameter Configuration=HP-search2026.0463.36
InstructBLIPBackbone=Vicuna-13B2023.1163.1
HalluVL-DPOwBackbone=LLaVA-1.5-7B2026.0463.01
Avg. of Single ModelsParam. Scale=Small2026.0362.95
HalluVL-DPOBackbone=LLaVA-1.5-7B2026.0462.71
Mini-Gemini (MGM-2B)#Tokens=18432, Compression Ratio (CR)=100%, Training Time=18.1h2024.0662.7
VCDBackbone=LLaVA-1.5-7B2026.0462.67
Mistral-7B (base)Backbone=Mistral-7B, Evaluation Protocol=Base2026.0362.24
LLaVolta (MGM-2B backbone)#Tokens=10863, Compression Ratio (CR)=170%, Training Time=14.8h2024.0662.2