List your company

Document Question Answering on SlideVQA (test)

63.2EMEagle-2.5
19.72831.01442.353.586Jan 17, 2026Jan 20, 2026Jan 23, 2026Jan 26, 2026Jan 29, 2026Feb 1, 2026Feb 4, 2026

Evaluation results

MethodLinks
Eagle-2.5Year=2025, Modal=V, Params=8B2026.0163.272.3—
AVIRModal=V, Params=3B2026.0160.368.9—
FRAGYear=2025, Modal=V, Params=7B2026.0159.865.1—
Qwen2.5-3B-AWQYear=2025, Modal=V, Params=3B2026.0156.665.8—
LILaC (w/ MM-Embed)MLLM=Qwen2.5-VL 7B, In-domain fine-tuned=false2026.0255.5766.32—
Arctic-TILTYear=2024, Modal=V, Params=0.8B2026.0155.1——
VisRAG-RetMLLM=Qwen2.5-VL 7B, In-domain fine-tuned=true2026.0255.0366.13—
NV-Embed-v2MLLM=Qwen2.5-VL 7B, In-domain fine-tuned=false2026.0253.7764.41—
ColPaliMLLM=Qwen2.5-VL 7B, In-domain fine-tuned=true2026.0253.7764.54—
LILaC (w/ UniME)MLLM=Qwen2.5-VL 7B, In-domain fine-tuned=false2026.0253.0562.89—
LILaC (w/ mmE5)MLLM=Qwen2.5-VL 7B, In-domain fine-tuned=false2026.0250.8959.07—
VisRAG-RetMLLM=MiniCPM V2.6, In-domain fine-tuned=true2026.0243.8862.37—
M3DYear=2023, Modal=TLV2026.0133.541.7—
InstructDrYear=2024, Modal=TLV, Params=3.4B2026.0131.940.2—
LayoutT5Year=2021, Modal=TLV2026.0131.739.9—
FIDYear=2020, Modal=T2026.0130.438.9—
T5Year=2020, Modal=T, Params=0.2B2026.0129.337.9—
BLIP-2Year=2023, Modal=TV, Params=3.4B2026.0128.338.8—
LayoutLMv2Year=2020, Modal=TLV2026.0121.429.3—
MARAModel=MiniCPM-V 2.62026.02——69.48
MARAModel=GPT-4o2026.02——73.4
MARA (Our Framework)Model=MiniCPM-V 2.6, Input=top-12026.02——62.68
MARA (Our Framework)Model=MiniCPM-V 2.6, Input=top-32026.02——66.4
MARA (Our Framework)Model=MiniCPM-V 2.6, Input=top-102026.02——67.56
MARA (Our Framework)Model=GPT-4o, Input=top-12026.02——67.79
MARA (Our Framework)Model=GPT-4o, Input=top-32026.02——69.2
OracleModel=MiniCPM-V 2.62026.02——73.66
OracleModel=GPT-4o2026.02——85.42
VDocRAGYear=2025, Modal=V, Params=8B2026.01—44.2—
VisRAGModel=MiniCPM-V 2.6, Input=top-12026.02——62.44
VisRAGModel=MiniCPM-V 2.6, Input=top-32026.02——67.87
VisRAGModel=MiniCPM-V 2.6, Input=top-102026.02——67.8
VisRAGModel=GPT-4o, Input=top-12026.02——64.74
VisRAGModel=GPT-4o, Input=top-32026.02——69.41