List your company

Visual Grounded Reasoning on TreeBench

54.8Overall Scoreo3-0416
39.40843.40447.451.396Jul 10, 2025Sep 4, 2025Oct 31, 2025Dec 27, 2025Feb 21, 2026Apr 19, 2026Jun 15, 2026

Evaluation results

MethodLinks
o3-0416Model Category=Private Model2025.1154.8—6969.265.268.879.422.438.66186.250——
GPT-o3-04162026.0354.8—6969.265.268.879.422.438.66186.250——
o3-0416Model Type=Private Models2025.0754.8—6969.265.268.879.422.438.66186.250——
o3-04162026.0254.8—6969.265.268.879.422.438.66186.250——
o3Parameters=—2026.0254.8—6969.265.268.879.422.438.66186.250——
Gemini-2.5-Pro-0605Model Category=Private Model2025.1154.1—51.761.556.57583.82036.865.986.254.6——
Gemini-2.5-Pro-06052026.0354.1—51.761.556.57583.82036.865.986.254.6——
Gemini-2.5-Pro-0605Model Type=Private Models2025.0754.1—51.761.556.57583.82036.865.986.254.6——
Gemini-2.5-Pro-06052026.0254.1—51.761.556.57583.82036.865.986.254.6——
Gemini-2.5-ProParameters=—2026.0254.1—51.761.556.57583.82036.865.986.254.6——
TTSPBackbone=Qwen3-VL-8B-Instruct2026.0451.8———————————73.239.4
Nemotron 3 Nano OmniOpen-Source=true, Size=30B-A3B, Reasoning=on2026.0451.6—————————————
GRiPModel Category=Open-source Visual Grounded Reasoning Model2025.1151.34569.154.483.469.364.123.238.658.769.445.8——
InternVL3-78BECRD=true2026.0250.9—65.569.265.27560.321.235.163.486.247.7——
TTSPBackbone=Qwen3-VL-4B-Instruct2026.0450.9———————————70.539.4
TreeVGR-7BModel Category=Open-source Visual Grounded Reasoning Model, Parameters=7B2025.1150.44465.553.882.668.863.322.436.8616945.5——
TreeVGR-7BModel Type=Open-source Visual Grounded Reasoning Models, Parameters=7B2025.0750.44465.553.882.668.863.322.436.8616945.5——
TreeVGR-7B2026.0250.4—65.553.882.668.863.322.436.8616945.5——
RTWIBackbone=Qwen3-VL-8B-Instruct2026.0450.1———————————71.137.9
Qwen2.5-VL-72BECRD=true2026.0249.9—65.569.265.262.569.12036.858.575.940.9——
DeepconfBackbone=Qwen3-VL-8B-Instruct2026.0449.9———————————70.437.9
TreeVGRPixel Constraint=640x28x28, Post-training Category=Non-efficiency-oriented Post Training Methods, Inference Pipeline=inference with original pipeline2025.0949.1—————————————
RTWIBackbone=Qwen3-VL-4B-Instruct2026.0449.1———————————66.439
SCBackbone=Qwen3-VL-8B-Instruct2026.0449.1———————————69.837.1
TRTBackbone=Qwen3-VL-8B-Instruct2026.0448.9———————————69.137.1
Qwen2.5-VL-32BECRD=true2026.0248.6—62.153.873.962.560.323.535.16165.545.5——
Self-Ref.Backbone=Qwen3-VL-8B-Instruct2026.0448.6———————————69.136.7
InternVL3-38BECRD=true2026.0248.4—51.769.265.27558.822.436.856.16950——
TreeVGRPixel Constraint=1280x28x28, Post-training Category=Non-efficiency-oriented Post Training Methods, Inference Pipeline=inference with original pipeline2025.0948.2—————————————
DeepconfBackbone=Qwen3-VL-4B-Instruct2026.0448.2———————————65.837.9
CISCBackbone=Qwen3-VL-8B-Instruct2026.0448.2———————————67.836.7
Self-Cer.Backbone=Qwen3-VL-8B-Instruct2026.0448.2———————————67.836.7
Qwen2.5-VL-7BECRD=true2026.0247.9—72.453.873.962.554.42035.156.175.945.5——
SCBackbone=Qwen3-VL-4B-Instruct2026.0447.7———————————64.437.9
Self-Ref.Backbone=Qwen3-VL-4B-Instruct2026.0447.7———————————64.437.9
CISCBackbone=Qwen3-VL-4B-Instruct2026.0447.2———————————63.837.5
TRTBackbone=Qwen3-VL-4B-Instruct2026.0447.2———————————63.837.5
Self-Cer.Backbone=Qwen3-VL-4B-Instruct2026.0447.2———————————63.737.5
GPT-4o-1120Model Category=Private Model2025.1146.9—51.761.565.243.869.118.838.648.872.443.2——
GPT-4o-11202026.0346.9—51.761.565.243.869.118.838.648.872.443.2——
GPT-4o-1120Model Type=Private Models2025.0746.9—51.761.565.243.869.118.838.648.872.443.2——
GPT-4o-11202026.0246.9—51.761.565.243.869.118.838.648.872.443.2——
LLaVA-OneVision-72BECRD=true2026.0246.9—62.161.569.668.851.518.829.86172.452.3——
GPT-4oParameters=—2026.0246.9—51.761.565.243.869.118.838.648.872.443.2——
GPT-4o2026.0446.9———————————61.738.3
GPT-4oVisual Bandwidth Constraint (B)=No Constrain, Reported From=Other publications2026.0546.9—————————————
GPT-4o2026.0646.9—51.761.565.243.869.118.838.648.872.443.2——
InternVL3-78BModel Category=Open-source General Model, Parameters=78B2025.1146.4—62.161.552.268.852.916.533.36186.245.5——
InternVL3-78BParameter Scale=78B2026.0346.4—62.161.552.268.852.916.533.36186.245.5——
InternVL3-78BModel Type=Open-source General Models, Parameters=78B2025.0746.4—62.161.552.268.852.916.533.36186.245.5——
InternVL3-78BECRD=false2026.0246.4—62.161.552.268.852.916.533.36186.245.5——
BaseBackbone=Qwen3-VL-8B-Instruct2026.0446.4———————————67.134.4
Gemini-2.5-Flash-0520Model Category=Private Model2025.1145.9—48.353.969.668.87515.319.356.172.443.2——
Gemini-2.5-Flash-05202026.0345.9—48.353.969.668.87515.319.356.172.443.2——
Gemini-2.5-Flash-0520Model Type=Private Models2025.0745.9—48.353.969.668.87515.319.356.172.443.2——
Gemini-2.5-Flash-05202026.0245.9—48.353.969.668.87515.319.356.172.443.2——
Gemini-2.5-FlashParameters=—2026.0245.9—48.353.969.668.87515.319.356.172.443.2——
InternVL3-8BECRD=true2026.0245.2—51.769.269.662.55024.729.843.972.450——
BaseBackbone=Qwen3-VL-4B-Instruct2026.0445.2———————————60.436.3
HART-7BParameters=7B2026.0243.7—62.161.552.256.355.914.135.148.862.150——
Nemotron 3 Nano OmniOpen-Source=true, Size=30B-A3B, Reasoning=off2026.0443.7—————————————
LLaVA-OneVision-7BECRD=true2026.0243.5—55.261.560.956.347.123.52853.772.440.9——
LazyMCoTBackbone=Qwen3-VL-8B-Instruct2026.0643.5—55.261.565.27545.616.529.848.872.450——
ERGOPixel Constraint=640x28x28, Post-training Category=Efficiency-oriented Post Training Methods2025.0943.2—————————————
BaseBackbone=Qwen3-VL-8B-Instruct2026.0643—55.261.569.67545.618.828.141.56950——
Qwen2.5-VL-32BModel Category=Open-source General Model, Parameters=32B2025.1142.5—51.753.869.662.554.416.533.346.362.138.6——
Qwen2.5-VL-32BParameter Scale=32B2026.0342.5—51.753.869.662.554.416.533.346.362.138.6——
DeepScanRL-based=false, Backbone=Qwen2.5-VL-7B2026.0342.537.362.169.260.968.844.121.236.843.948.343.2——
Qwen2.5-VL-32BECRD=false2026.0242.5—51.753.869.662.554.416.533.346.362.138.6——
Qwen2.5-VL-32BParameters=32B2026.0242.5—51.753.869.662.554.416.533.346.362.138.6——
Nemotron Nano V2 VLOpen-Source=true, Size=12B, Reasoning=on2026.0442.5—————————————
PixelReasonerPixel Constraint=640x28x28, Post-training Category=Non-efficiency-oriented Post Training Methods2025.0942.3—————————————
Qwen2.5-VL-72BModel Category=Open-source General Model, Parameters=72B2025.1142.2—65.569.256.556.348.511.833.351.272.438.6——
Qwen2.5-VL-72BParameter Scale=72B2026.0342.2—65.569.256.556.348.511.833.351.272.438.6——
Qwen2.5-VL-72BModel Type=Open-source General Models, Parameters=72B2025.0742.2—65.569.256.556.348.511.833.351.272.438.6——
Qwen2.5-VL-72BECRD=false2026.0242.2—65.569.256.556.348.511.833.351.272.438.6——
Qwen2.5-VL-72BParameters=72B2026.0242.2—65.569.256.556.348.511.833.351.272.438.6——
InternVL3-38BModel Category=Open-source General Model, Parameters=38B2025.1142—51.761.552.268.851.512.933.356.165.538.6——
InternVL3-38BParameter Scale=38B2026.0342—51.761.552.268.851.512.933.356.165.538.6——
InternVL3-38BECRD=false2026.0242—51.761.552.268.851.512.933.356.165.538.6——
InternVL3-38BParameters=38B2026.0242—51.761.552.268.851.512.933.356.165.538.6——
MGPOPixel Constraint=640x28x28, Post-training Category=Efficiency-oriented Post Training Methods, Inference Pipeline=reproduction with their code using our data2025.0942—————————————
ERGOPixel Constraint=1280x28x28, Post-training Category=Efficiency-oriented Post Training Methods2025.0941.7—————————————
LazyMCoTBackbone=Qwen2.5-VL-7B-Instruct2026.0641.7—58.661.556.568.839.721.231.646.348.354.5——
HiDeBackbone=Qwen3-VL-8B-Instruct2026.0641.7—48.346.265.27545.615.324.653.765.552.3——
TreeVGRRL-based=true, Backbone=Qwen2.5-VL-7B2026.034131.855.261.565.25061.715.324.646.344.840.9——
VisionThinkPixel Constraint=1280x28x28, Post-training Category=Efficiency-oriented Post Training Methods, Inference Pipeline=inference with original pipeline2025.0941—————————————
CVSearchBackbone=Qwen2.5-VL-7B2026.0540.7—————————————
LLaVA-OneVision-72BModel Category=Open-source General Model, Parameters=72B2025.1140.5—62.153.865.262.336.812.928.153.765.547.7——
LLaVA-OneVision-72BParameter Scale=72B2026.0340.5—62.153.865.262.336.812.928.153.765.547.7——
LLaVA-OneVision-72BModel Type=Open-source General Models, Parameters=72B2025.0740.5—62.153.865.262.336.812.928.153.765.547.7——
LLaVA-OneVision-72BECRD=false2026.0240.5—62.153.865.262.536.812.928.153.765.547.7——
LLaVA-OneVision-72BParameters=72B2026.0240.5—62.153.865.262.336.812.928.153.765.547.7——
PixelReasonerPixel Constraint=1280x28x28, Post-training Category=Non-efficiency-oriented Post Training Methods2025.0940.4—————————————
MGPO + SFTParameters=7B2026.0240.3—58.661.552.256.348.514.131.651.251.740.9——
Qwen2.5-VL + BA-SFT + VanillaRLVisual Bandwidth Constraint (B)=No Constrain, Tool-calling=True2026.0540.25—————————————
Qwen2.5-VL + BA-SFTVisual Bandwidth Constraint (B)=No Constrain, Tool-calling=True2026.0540.24—————————————
DeepEyesPixel Constraint=1280x28x28, Post-training Category=Non-efficiency-oriented Post Training Methods2025.0940.2—————————————
Qwen2.5-VL-7B-Inst.Pixel Constraint=640x28x282025.0940.2—————————————
Qwen2.5-VL-7B-Inst.Pixel Constraint=16384x28x282025.0940—————————————