List your company
ResearchFormerly WizWand

Diagram Question Answering on AI2D (test)

94.7AccuracyClaude 3.5
72.13277.99183.8589.709Jun 24, 2024Oct 5, 2024Jan 16, 2025Apr 29, 2025Aug 10, 2025Nov 21, 2025Mar 4, 2026
Updated 8mo ago

Evaluation results

MethodLinks
Claude 3.5evaluation_protocol=transparent bounding box2024.0794.7
Claude 3.5 Sonnet2024.1294.7
Gemini 1.5 Proevaluation_protocol=transparent bounding box2024.0794.4
Gemini-1.5-Pro2024.1294.4
Gemini 1.5 Pro2024.1294.4
GPT-4oevaluation_protocol=transparent bounding box2024.0794.2
GPT-4o2024.1294.2
GPT-4o2024.1294.2
NVLM-D-1.0# Parameters=78B2024.1294.2
Llama 3-V 405Bevaluation_protocol=transparent bounding box2024.0794.1
NVILA# Parameters=15B2024.1294.1
GenRecalstudent=Qwen2-VL-7B, teacher=InternVL2.5-78B2025.0693.9
Llama 3-V 70Bevaluation_protocol=transparent bounding box2024.0793
GenRecalstudent=InternVL2.5-8B, teacher=InternVL2.5-78B2025.0693
NVILA# Parameters=8B2024.1292.3
Llama 3.2# Parameters=90B2024.1292.3
NVILA-Lite# Parameters=15B2024.1292
Llama 3.2# Parameters=11B2024.1291.9
GenRecalstudent=InternVL2.5-2B, teacher=InternVL2.5-78B2025.0691.8
NVILA-Lite# Parameters=8B2024.1291
GenRecalstudent=InternVL2.5-4B, teacher=InternVL2.5-78B2025.0690
GenRecalstudent=Qwen2-VL-2B, teacher=InternVL2.5-78B2025.0689.1
VILA-1.5# Parameters=40B2024.1288.9
InternVL2# Parameters=40B2024.1287.1
LLaVA-OneVision# Parameters=72B2024.1285.6
Qwen3-VL-32B-InstructContext Window=32K2026.0385
InternVL2.5size=8B2025.0684.8
Phi-4-reasoning-vision-15B2026.0384.8
Qwen3-VL-32B-InstructContext Window=4K2026.0384.8
Phi-4-reasoning-vision-15BProtocol=force nothink2026.0384.7
GPT-4oZero-shot=true, Decoding Strategy=greedy2024.0984.6
Kimi-VL-A3B-Instruct2026.0384.6
Llama 3-V 8Bevaluation_protocol=transparent bounding box2024.0784.4
Qwen2.5VLSize=7B, Type=AR, Samples=>9M2025.1283.9
InternVL2Size=8B, #token /image tile=2562024.1283.8
PVC InternVL2 (Ours)Size=8B, #token /image tile=642024.1283.8
InternVL2# Parameters=8B2024.1283.8
InternVL2size=8B2025.0683.8
Qwen2-VL-InstructLLM=Qwen2-7B, Vision Encoder=DFN-CLIP-H2024.1283
Qwen2-VLSize=7B, #token /image tile=Native resolution (unfixed)2024.1283
Qwen2-VL# Parameters=8B2024.1283
InternVL-2-26BModel=InternVL-2-26B2026.0183
Qwen3-VL-8B-InstructContext Window=32K2026.0383
GenRecalstudent=InternVL2.5-1B, teacher=InternVL2.5-78B2025.0682.9
Qwen3-VL-8B-InstructContext Window=4K2026.0382.7
DiffusionVLSize=7B, Type=Diff., Samples=738K2025.1282.2
InternVL-2-8BModel=InternVL-2-8B2026.0182.2
MiniCPM-V2.6size=8B2025.0682.1
LLaVA-OV (SI)LLM=Qwen2-7B, Vision Encoder=SigLIP-SO400M2024.1281.6
Qwen2.5VLSize=3B, Type=AR, Samples=>9M2025.1281.6
IXC-2.5Size=7B, #token /image tile=4002024.1281.5
LLaVA-OVLLM=Qwen2-7B, Vision Encoder=SigLIP-SO400M2024.1281.4
LLaVA-OVSize=7B, #token /image tile=7292024.1281.4
LLaVA-OneVision# Parameters=8B2024.1281.4
LLaVA-OVSize=7B, Type=AR, Samples=7.8M2025.1281.4
LLaVA-OneVisionsize=7B2025.0681.4
InternVL2.5size=4B2025.0681.4
gemma-3-12b-it2026.0380.4
VILA-1.5# Parameters=13B2024.1279.9
Cambrian-34BModel Scale=30B, Zero-shot=true, Decoding Strategy=greedy2024.0979.7
Cambrian-1# Parameters=34B2024.1279.7
Gemini-1.5-ProZero-shot=true, Decoding Strategy=greedy2024.0979.1
Pixtral# Parameters=12B2024.1279
InternVL2size=4B2025.0678.9
LLaVA-NeXT-INST-ITLLM=Qwen2-7B, Vision Encoder=SigLIP-SO4002024.1278.7
Oryx MLLMSize=7B2024.1278.5
DiffusionVLSize=3B, Type=Diff., Samples=738K2025.1278.4
MiniCPM-V2.5size=8B2025.0678.4
GPT-4Vevaluation_protocol=transparent bounding box2024.0778.2
GPT-4V2024.1278.2
LLaVA-OV-7B w/ CLIModel=LLaVA-OV-7B w/ CLI2026.0177.9
LLaDA-VSize=8B, Type=Diff., Samples=16.5M2025.1277.8
Phi-3.5-Visionsize=4B2025.0677.8
LLaVA-OV-7B w/ SLIModel=LLaVA-OV-7B w/ SLI2026.0177.6
LLaVA-OV-7BModel=LLaVA-OV-7B2026.0177.5
Qwen2-VLsize=7B2025.0677.5
MM1.5-30BModel Scale=30B, Zero-shot=true, Decoding Strategy=greedy2024.0977.2
Phi-3-Vision-4BModel Scale=3B, Zero-shot=true, Decoding Strategy=greedy2024.0976.7
VILA-1.5# Parameters=8B2024.1276.6
PVC InternVL2 (Ours)Size=2B, #token /image tile=642024.1276.3
Eaglesize=8B2025.0676.1
LLaVA-OV-7B w/ DeepStackModel=LLaVA-OV-7B w/ DeepStack2026.0176
GPT-4VZero-shot=true, Decoding Strategy=greedy2024.0975.9
Aquila-VLsize=2B2025.0675
InternVL2.5size=2B2025.0674.9
Qwen2-VLSize=2B, #token /image tile=Native resolution (unfixed)2024.1274.7
DimpleSize=7B, Type=Diff., Samples=1.3M2025.1274.4
Florence-VL# Parameters=8B2024.1274.2
InternVL2-2BModel Scale=1B, Zero-shot=true, Decoding Strategy=beam search2024.0974.1
InternVL2Size=2B, #token /image tile=2562024.1274.1
InternVL2size=2B2025.0674.1
Eaglesize=13B2025.0674
Cambrian-1# Parameters=13B2024.1273.6
Cambrian-1size=13B2025.0673.6
LLaVA-NeXT-Qwen22024.0673.5
CogVLM2size=19B2025.0673.4
MM1-30BModel Scale=30B, Zero-shot=true, Decoding Strategy=greedy2024.0973.3
LongVAResolution Strategy=AnyRes2024.0673.1
Cambrian-1# Parameters=8B2024.1273
Cambrian-1Size=8B, Type=AR, Samples=-2025.1273