List your company

Document Question Answering on DocVQA (test)

96.4AccuracyQwen2.5-VL-72B
25.36843.80962.2580.691Jul 31, 2024Nov 9, 2024Feb 18, 2025May 30, 2025Sep 8, 2025Dec 18, 2025Mar 30, 2026

Evaluation results

MethodLinks
Qwen2.5-VL-72BModel Size=72B2025.0296.4
Qwen3-VL-8BAccess=Open weights only2026.0396.1
Qwen2.5-VL-7BModel Size=7B2025.0295.7
Qwen2.5-VL-7BJudge=GPT-4o2025.1295.5
MiMo-VL-7B-SFT-2508Judge=GPT-4o2025.1295.5
Qwen3-VL-4BAccess=Open weights only2026.0395.3
Claude 3.52024.0795.2
Claude-3.5 Sonnet2025.0295.2
MiMo-VL-Miloco-7BJudge=GPT-4o2025.1295.2
InternVL2.5-78BModel Size=78B2025.0295.1
MiniCPM-V-4.5-8BAccess=Open weights only2026.0394.7
PLM-8BAccess=Fully open2026.0394.6
Qwen2-VLSize=7B, #token /image tile=Native resolution (unfixed)2024.1294.5
Qwen2-VL# Parameters=8B2024.1294.5
Eagle2.5-8BAccess=Open weights only2026.0394.1
NVILA# Parameters=15B2024.1294
Qwen2.5-VL-3BModel Size=3B2025.0293.9
InternVL2# Parameters=40B2024.1293.9
PLM-3BAccess=Fully open2026.0393.8
MolmoPoint-8BAccess=MolmoPoint2026.0393.8
NVILA# Parameters=8B2024.1293.7
Keye-VL-1.5-8BAccess=Open weights only2026.0393.4
GLM-4.1V-9BAccess=Open weights only2026.0393.3
Molmo2-8BAccess=Fully open2026.0393.2
Gemini 1.5 Proexternal_ocr_tools=true2024.0793.1
Gemini-1.5-Pro2024.1293.1
Gemini 1.5 Pro2025.0293.1
Gemini 1.5 Pro2024.1293.1
GPT-4o2024.0792.8
GPT-4o2024.1292.8
GPT-4o2024.1292.8
Llama 3-V 405B2024.0792.6
NVLM-D-1.0# Parameters=78B2024.1292.6
PVC InternVL2 (Ours)Size=8B, #token /image tile=642024.1292.5
InternVL3.5-4BAccess=Open weights only2026.0392.4
InternVL3.5-8BAccess=Open weights only2026.0392.3
Llama 3-V 70B2024.0792.2
NVILA-Lite# Parameters=8B2024.1291.7
Claude Sonnet 4.5Access=API call only2026.0391.7
InternVL2Size=8B, #token /image tile=2562024.1291.6
InternVL2# Parameters=8B2024.1291.6
Gemini 2.5 ProAccess=API call only2026.0391.5
LLaVA-OneVision# Parameters=72B2024.1291.3
GPT-4o2025.0291.1
Gemini 2.5 FlashAccess=API call only2026.0391.1
IXC-2.5Size=7B, #token /image tile=4002024.1290.9
Pixtral# Parameters=12B2024.1290.7
NVILA-Lite# Parameters=15B2024.1290.6
MolmoPoint-8B-O-7BAccess=Fully open2026.0390.4
Qwen2-VLSize=2B, #token /image tile=Native resolution (unfixed)2024.1290.1
Llama 3.2# Parameters=90B2024.1290.1
OracleModel=MiniCPM-V 2.62026.0289.41
InternVL3-8BJudge=GPT-4o2025.1289.4
Oryx MLLMSize=7B2024.1289
GPT-5Access=API call only2026.0388.9
GPT-4V2024.0788.4
GPT-4V2024.1288.4
Llama 3.2# Parameters=11B2024.1288.4
GPT-4VModel Category=Closed-source Models2025.1288.4
SDAR-VL-8B-InstModel Category=Block-wise Discrete Diffusion Models2025.1288.3
OracleModel=GPT-4o2026.0288.1
Molmo2-4BAccess=Fully open2026.0387.8
PVC InternVL2 (Ours)Size=2B, #token /image tile=642024.1287.7
LLaVA-OVSize=7B, #token /image tile=7292024.1287.5
LLaVA-OneVision# Parameters=8B2024.1287.5
InternVL2Size=2B, #token /image tile=2562024.1286.9
GPT-5 miniAccess=API call only2026.0386.7
Gemma-3-27B-ITJudge=GPT-4o2025.1286.6
Claude 3.5 Sonnet2024.1285.2
Florence-VL# Parameters=8B2024.1284.9
MARAModel=GPT-4o2026.0284.64
Llama 3-V 8B2024.0784.4
MARA (Our Framework)Model=GPT-4o, Input=top-32026.0281.22
LongVASize=7B, #token /image tile=1442024.1280.8
MARAModel=MiniCPM-V 2.62026.0280.59
VisRAGModel=GPT-4o, Input=top-32026.0278.21
Cambrian-1# Parameters=8B2024.1277.8
Cambrian-1# Parameters=13B2024.1276.8
MARA (Our Framework)Model=MiniCPM-V 2.6, Input=top-32026.0276.53
Cambrian-1# Parameters=34B2024.1275.5
VisRAGModel=MiniCPM-V 2.6, Input=top-32026.0274.5
MARA (Our Framework)Model=GPT-4o, Input=top-12026.0270.28
MARA (Our Framework)Model=MiniCPM-V 2.6, Input=top-12026.0269.45
VisRAGModel=MiniCPM-V 2.6, Input=top-12026.0269.4
VisRAGModel=GPT-4o, Input=top-12026.0268.32
MARA (Our Framework)Model=MiniCPM-V 2.6, Input=top-102026.0260.56
VisRAGModel=MiniCPM-V 2.6, Input=top-102026.0259.07
VILA-1.5# Parameters=13B2024.1258.6
VILA-1.5# Parameters=40B2024.1258.6
VILA-1.5# Parameters=8B2024.1240.6
LLaVA-1.5# Parameters=13B2024.1230.3
LLaVA-1.5# Parameters=7B2024.1228.1