Visual Search on V* benchmark
91.1Overall Success RateRAP
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| RAPexternal_tool=true2026.05 | 91.1 | 90.4 | 96.1 | |
| SeProDbase_method=DeepEyes, external_tool=false, prophet_model_size=3B2026.05 | 91.1 | 91.3 | 90.8 | |
| PFlowNetModel Category=Visually Grounded Reasoning Models2026.05 | 90.6 | 91.4 | 89.5 | |
| ZoomEyesexternal_tool=true2026.05 | 90.6 | 93.9 | 85.5 | |
| FOCUSexternal_tool=true2026.05 | 90.6 | — | — | |
| Simple-o3external_tool=false2026.05 | 90.4 | — | — | |
| SeProDbase_method=Mini-o3, external_tool=false, prophet_model_size=7B2026.05 | 90.4 | 92 | 88 | |
| DeepEyesModel Category=Visually Grounded Reasoning Models2026.05 | 90 | 92.1 | 86.8 | |
| SeProDbase_method=Mini-o3, external_tool=false, prophet_model_size=3B2026.05 | 89.2 | 90.8 | 86.8 | |
| DeepEyesexternal_tool=false2026.05 | 89 | 90.4 | 86.8 | |
| SeProDbase_method=Pixel Reasoner, external_tool=false, prophet_model_size=3B2026.05 | 88.5 | 90.4 | 85.5 | |
| Chain-of-Focusexternal_tool=false2026.05 | 88 | — | — | |
| Qwen3-VL-32BModel Category=General Large Vision-Language Models2026.05 | 87.4 | 87 | 88.2 | |
| TreeVGRModel Category=Visually Grounded Reasoning Models2026.05 | 87.4 | 89.5 | 84.2 | |
| Pixel Reasonerexternal_tool=false2026.05 | 86.9 | 88.7 | 84.2 | |
| ChainOfFocusVisual Bandwidth Constraint (B)=No Constrain, Tool-calling=True2026.05 | 86.39 | 86.09 | 86.84 | |
| Mini-o3external_tool=false2026.05 | 86.3 | 88.8 | 82.4 | |
| Qwen2.5-VL-32BModel Category=General Large Vision-Language Models2026.05 | 85.9 | 83.5 | 89.5 | |
| Qwen2.5-VL + BA-SFT + VanillaRLVisual Bandwidth Constraint (B)=No Constrain, Tool-calling=True2026.05 | 85.86 | 87.83 | 82.89 | |
| Starve to PerceiveVisual Bandwidth Constraint (B)=No Constrain, Tool-calling=True2026.05 | 85.86 | 86.96 | 84.21 | |
| Qwen2.5-VL-72BModel Category=General Large Vision-Language Models2026.05 | 84.8 | 90.8 | 80.9 | |
| DeepEyesVisual Bandwidth Constraint (B)=No Constrain, Tool-calling=True2026.05 | 83.77 | 82.61 | 85.53 | |
| PixelReasonerVisual Bandwidth Constraint (B)=No Constrain, Tool-calling=True2026.05 | 83.25 | 85.22 | 80.26 | |
| ThymeModel Category=Visually Grounded Reasoning Models2026.05 | 82.2 | 83.5 | 80.3 | |
| DeepEyesV2Model Category=Visually Grounded Reasoning Models2026.05 | 81.8 | 81.7 | 80.3 | |
| DyFoexternal_tool=true2026.05 | 81.2 | 80 | 82.9 | |
| Qwen2.5-VL + BA-SFTVisual Bandwidth Constraint (B)=No Constrain, Tool-calling=True2026.05 | 80.62 | 85.22 | 73.68 | |
| PixelReasonerModel Category=Visually Grounded Reasoning Models2026.05 | 80.6 | 83.5 | 76.3 | |
| Gemini-2.5-ProVisual Bandwidth Constraint (B)=No Constrain, Reported From=Other publications2026.05 | 79.2 | — | — | |
| InternVL3-38BModel Category=General Large Vision-Language Models2026.05 | 77.5 | 77.4 | 77.6 | |
| Qwen3-VL-8BModel Category=General Large Vision-Language Models2026.05 | 77.5 | 80.2 | 73.7 | |
| LLaVA-OneVisionVisual Bandwidth Constraint (B)=No Constrain, Reported From=Other publications2026.05 | 75.4 | 75.7 | 75 | |
| SEALexternal_tool=true2026.05 | 75.4 | 74.8 | 76.3 | |
| SEALVisual Search Mechanism=yes2026.02 | 75.39 | 74.78 | 76.31 | |
| Mini-o3Visual Bandwidth Constraint (B)=No Constrain, Tool-calling=True2026.05 | 75.39 | 72.17 | 80.26 | |
| Qwen3-VL-4BModel Category=General Large Vision-Language Models2026.05 | 74.9 | 78.3 | 69.7 | |
| Qwen2.5-VL-7BModel Category=General Large Vision-Language Models2026.05 | 74.3 | 77.4 | 69.7 | |
| LLaVA-OV-72BModel Category=General Large Vision-Language Models2026.05 | 73.8 | 80.9 | 63.2 | |
| RegionReasoner-7BVisual Search Mechanism=no2026.02 | 72.77 | 75.65 | 68.42 | |
| InternVL3-8BModel Category=General Large Vision-Language Models2026.05 | 72.3 | 73 | 71.1 | |
| Qwen2.5-VLVisual Bandwidth Constraint (B)=No Constrain, Reported From=Other publications2026.05 | 71.2 | 73.9 | 67.1 | |
| VisionReasoner-7BVisual Search Mechanism=no2026.02 | 69.63 | 75.62 | 60.52 | |
| Qwen2.5-VL + BA-SFT + VanillaRLVisual Bandwidth Constraint (B)=256, Tool-calling=True2026.05 | 67.54 | 66.09 | 69.74 | |
| Qwen2.5-VL-7BVisual Search Mechanism=no2026.02 | 67.53 | 72.17 | 60.52 | |
| Starve to PerceiveVisual Bandwidth Constraint (B)=256, Tool-calling=True2026.05 | 66.49 | 62.61 | 72.36 | |
| GPT-4o-1120Model Category=General Large Vision-Language Models2026.05 | 66 | — | — | |
| GPT-4oVisual Bandwidth Constraint (B)=No Constrain, Reported From=Other publications2026.05 | 62.8 | — | — | |
| DeepEyesVisual Bandwidth Constraint (B)=256, Tool-calling=True2026.05 | 62.3 | 60.87 | 64.47 | |
| ChainOfFocusVisual Bandwidth Constraint (B)=256, Tool-calling=True2026.05 | 60.7 | 55.7 | 68.4 | |
| Qwen2.5-VL + BA-SFTVisual Bandwidth Constraint (B)=256, Tool-calling=True2026.05 | 59.16 | 55.65 | 64.47 | |
| Mini-o3Visual Bandwidth Constraint (B)=256, Tool-calling=True2026.05 | 57.59 | 53.04 | 64.47 | |
| GPT-4VVisual Search Mechanism=no2026.02 | 54.97 | 51.3 | 60.52 | |
| PixelReasonerVisual Bandwidth Constraint (B)=256, Tool-calling=True2026.05 | 54.45 | 46.96 | 65.79 | |
| Qwen2.5-VLVisual Bandwidth Constraint (B)=2562026.05 | 29.31 | 19.13 | 44.74 |