List your company
ResearchFormerly WizWand

Science Question Answering on SciQA-IMG

89SciQA-IMG AccuracyPhi 3.5 Vision
38.14451.34764.5577.753May 11, 2023Oct 24, 2023Apr 7, 2024Sep 20, 2024Mar 5, 2025Aug 18, 2025Jan 31, 2026
Updated 8mo ago

Evaluation results

MethodLinks
Phi 3.5 Vision2024.1289
Florence-VL 8B# Vis tok.=5762024.1285.9
Florence-VL 3B# Vis tok.=5762024.1284.6
SEA-PRIMELLM=Vicuna-13B, Res.=3842024.0880.9
Cambrian 8B# Vis tok.=5762024.1280.4
Vila 8B2024.1279.9
SEA-PRIMELLM=Llama3-8B, Res.=3842024.0879
SEA-PRIMELLM=Phi3-3.8B, Res.=3842024.0878.7
Mini-Gemini-HD 8B# Vis tok.=28802024.1275.1
SEA-PRIMELLM=Vicuna-7B, Res.=3842024.0873.9
SparseCutLLM=Vicuna-13B, Eff. Res.=672, Vis. Len.=2880, Con. Len.=576, PT=558K, SFT=665K2026.0171.8
LLaVA-1.5*LLM=Vicuna-13B, Res.=3362024.0871.6
LLaVALLM=Vicuna-13B, Eff. Res.=336, Vis. Len.=576, Con. Len.=576, PT=558K, SFT=665K2026.0171.6
InstructBLIPzero-shot=true, backbone=FlanT5-XXL2023.0570.6
LLaVA-1.5 + HALVABackbone=LLaVA-1.5-7B, Learning Objective=HALVA2025.0670.5
InstructBLIPzero-shot=true, backbone=FlanT5-XL2023.0570.4
LLaVA-1.5 + HA-DPOBackbone=LLaVA-1.5-7B, Learning Objective=HA-DPO2025.0670.3
LLaVA-1.5 + TL-DPOBackbone=LLaVA-1.5-7B, Learning Objective=TL-DPO2025.0670.3
AlignGPTLLM=Vicuna-13B, Res.=3362024.0870.3
MobileVLM-V2-3BLLM=MLLAMA 2.7B, Res.=3362024.0870
LLaMA-VIDLLM=Vicuna-13B, Res.=3362024.0870
SparseCutLLM=Vicuna-7B, Eff. Res.=672, Vis. Len.=2880, Con. Len.=576, PT=558K, SFT=665K2026.0170
Visual PromptLLM=Vicuna-7B, Res.=3362024.0869.5
SEA-PRIMELLM=Gemma-2B, Res.=3842024.0869.2
TinyLLaVALLM=Phi-2.7B, Res.=3842024.0869.1
AlignGPTLLM=Vicuna-7B, Res.=3362024.0868.5
LLaVA-PhiLLM=Phi-2.7B, Res.=3362024.0868.4
ShareGPT4VLLM=Vicuna-7B, Res.=3362024.0868.4
LLaMA-VIDLLM=Vicuna-7B, Res.=3362024.0868.3
Qwen-VL-ChatLLM=Qwen-7B, Res.=4482024.0868.2
LLaVA-1.5*LLM=Vicuna-7B, Res.=3362024.0867.9
Vila 3B2024.1267.9
LLaVA-1.5 + POVIDBackbone=LLaVA-1.5-7B, Learning Objective=POVID2025.0667.5
DeepStack-LLLM=Vicuna-7B, Eff. Res.=672, Vis. Len.=2880, Con. Len.=576, PT=558K, SFT=665K2026.0167.2
LLaVA-1.5 + RLHF-VBackbone=LLaVA-1.5-7B, Learning Objective=RLHF-V2025.0667.1
Qwen-VLLLM=Qwen-7B, Res.=4482024.0867.1
Qwen-VLLLM=Qwen-7B, Eff. Res.=448, Vis. Len.=256, Con. Len.=256, PT=1.4B, SFT=50M2026.0167.1
LLaVA-1.5Backbone=LLaVA-1.5-7B2025.0666.8
LLaVALLM=Vicuna-7B, Eff. Res.=336, Vis. Len.=576, Con. Len.=576, PT=558K, SFT=665K2026.0166.8
LLaVA-1.5 + VLfeedbackBackbone=LLaVA-1.5-7B, Learning Objective=VLfeedback2025.0666.2
LLaVA-1.5 + Human-PreferenceBackbone=LLaVA-1.5-7B, Learning Objective=Human-Preference2025.0665.8
BLIP-2zero-shot=true, backbone=FlanT5-XXL2023.0564.5
InstructBLIPLLM=Vicuna-13B, Res.=2242024.0863.1
InstructBLIPzero-shot=true, backbone=Vicuna-13B2023.0563.1
InstructBLIPLLM=Vicuna-13B, Eff. Res.=224, Vis. Len.=32, Con. Len.=32, PT=129M, SFT=1.2M2026.0163.1
MobileVLM-3BLLM=MLLAMA 2.7B, Res.=3362024.0861
BLIP-2zero-shot=true, backbone=Vicuna-13B2023.0561
BLIP-2LLM=Vicuna-13B, Eff. Res.=224, Vis. Len.=32, Con. Len.=32, PT=129M, SFT=-2026.0161
InstructBLIPzero-shot=true, backbone=Vicuna-7B2023.0560.5
InstructBLIPLLM=Vicuna-7B, Eff. Res.=224, Vis. Len.=32, Con. Len.=32, PT=129M, SFT=1.2M2026.0160.5
BLIP-2zero-shot=true, backbone=FlanT5-XL2023.0554.9
BLIP-2zero-shot=true, backbone=Vicuna-7B2023.0553.8
LLaVA next 8B# Vis tok.=28802024.1240.1