Visual Reasoning on Jigsaw
88.6AccuracyAdaReasoner 7B
Evaluation results
| Method | Links | |
|---|---|---|
| AdaReasoner 7BModel Category=Tool-Planning Models, Backbone=7B2026.01 | 88.6 | |
| GPT 5 + ToolsModel Category=Tool-Planning Models, Tool Usage=true2026.01 | 84.5 | |
| GPT 5Model Category=Closed-Source Models2026.01 | 80.1 | |
| CoLVR2026.05 | 78 | |
| GPT-4oVisual Sketchpad=Enabled2024.06 | 70.7 | |
| Qwen 2.5 VL 72BModel Category=Open-Source Models, Backbone=72B2026.01 | 70.1 | |
| GPT-4V-preview2024.06 | 70 | |
| SOTA [35]2024.06 | 70 | |
| Mirage2026.05 | 70 | |
| GPT-4 TurboVisual Sketchpad=Enabled2024.06 | 68.5 | |
| Qwen2.5-VL-7BTraining Paradigm=vanilla SFT + GRPO2026.05 | 68 | |
| DeepLatent-RL-7BModel=DeepLatent-RL-7B2026.05 | 68 | |
| Qwen2.5-VL-7BTraining Paradigm=vanilla SFT2026.05 | 67.33 | |
| Gemini 2.5 flashModel Category=Closed-Source Models2026.01 | 67.2 | |
| EVA-VL-7B2026.06 | 66.7 | |
| GPT-4 TurboVisual Sketchpad=Disabled2024.06 | 64.7 | |
| GPT-4oVisual Sketchpad=Disabled2024.06 | 64 | |
| Qwen 2.5 VL 72B + ToolsModel Category=Tool-Planning Models, Tool Usage=true, Backbone=72B2026.01 | 61.5 | |
| Qwen 2.5 VL 32BModel Category=Open-Source Models, Backbone=32B2026.01 | 59.5 | |
| Claude 4 sonnetModel Category=Closed-Source Models2026.01 | 58.6 | |
| LLaVA-1.5-13B2024.06 | 58 | |
| GPT-4o2026.06 | 58 | |
| Gemini-Pro2024.06 | 57.3 | |
| LLaVA-NeXT-34B2024.06 | 54.7 | |
| InternVL3 78BModel Category=Open-Source Models, Backbone=78B2026.01 | 52.8 | |
| Pixel-Reasoner-7BModel=Pixel-Reasoner-7B2026.05 | 52.7 | |
| LVR-7Bsource=collected from [23]2026.06 | 52.7 | |
| PixelReasonerModel Category=Tool-Planning Models, Unified Evaluation Framework=true2026.01 | 52.2 | |
| Qwen2.5-VL-7BModel=Qwen2.5-VL-7B2026.05 | 52 | |
| LVR-7BModel=LVR-7B2026.05 | 52 | |
| Qwen2.5-VL-7B2026.06 | 52 | |
| InternVL3-8BModel=InternVL3-8B2026.05 | 50 | |
| Monet-7BModel=Monet-7B2026.05 | 50 | |
| Qwen 2.5 VL 7BModel Category=Open-Source Models, Backbone=7B2026.01 | 45.7 | |
| Qwen 2.5 VL 7B + ToolsModel Category=Tool-Planning Models, Tool Usage=true, Backbone=7B2026.01 | 45 | |
| DeepEyesModel Category=Tool-Planning Models, Unified Evaluation Framework=true2026.01 | 44.8 | |
| Gemini-2.5-flash2026.05 | 43.33 | |
| Qwen 2.5 VL 3BModel Category=Open-Source Models, Backbone=3B2026.01 | 39.8 | |
| GPT-4o2026.05 | 36 | |
| Claude 3 OPUS2024.06 | 32.7 | |
| Qwen2.5-VL-7BReasoning Strategy=CoT2026.05 | 24.67 | |
| Qwen2.5-VL-7B2026.05 | 24 | |
| InternVL3.5-8B2026.05 | 19.33 | |
| LLaVA-1.5-7B2024.06 | 11.3 |