High-Resolution Fine-Grained Visual Perception on HR-Bench 4K
83.9Overall ScoreGemini-2.5-Pro
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Gemini-2.5-ProModel Category=Proprietary Model2026.06 | 83.9 | 85.5 | 82.3 | |
| Imagine-OPD-8BModel Category=Our Models, Backbone=8B2026.06 | 83.5 | 90.8 | 76.3 | |
| Imagine-OPD-4BModel Category=Our Models, Backbone=4B2026.06 | 83.3 | 90.3 | 76.3 | |
| Qwen3-VL-235B-A22BModel Category=Open-Source Models, Backbone=235B-A22B2026.06 | 83 | 89 | 77 | |
| Kimi-K2.5 (1T)Model Category=Open-Source Models, Backbone=1T2026.06 | 81.7 | 87.3 | 76 | |
| Qwen3-VL-8BModel Category=Open-Source Models, Backbone=8B2026.06 | 78.9 | 87.8 | 70 | |
| Qwen3-VL-30B-ThinkingModel Category=Open-Source Models, Backbone=30B2026.06 | 78.5 | 88.5 | 68.5 | |
| Qwen3-VL-4BModel Category=Open-Source Models, Backbone=4B2026.06 | 78.3 | 87.5 | 69.3 | |
| Gemini-2.5-FlashModel Category=Proprietary Model2026.06 | 77.5 | 81.5 | 74 | |
| Gemini-3-FlashModel Category=Proprietary Model2026.06 | 77.5 | 81.5 | 74 | |
| TreeVGR-7BModel Category=Reasoning with Tool Methods, Backbone=7B2026.06 | 77.1 | 90.3 | 64 | |
| ThymeModel Category=Reasoning with Tool Methods2026.06 | 77 | 91 | 63 | |
| DeepEyesModel Category=Reasoning with Tool Methods2026.06 | 75.1 | 91.3 | 59 | |
| Qwen2.5-VL-32BModel Category=Open-Source Models, Backbone=32B2026.06 | 73.4 | 87.5 | 59.3 | |
| Pixel-ReasonerModel Category=Reasoning with Tool Methods2026.06 | 72.9 | 86 | 60.3 | |
| AMVLAccess=Open-Source, Model size=7B2026.07 | 72.12 | 87.25 | 57 | |
| PixelReasonerAccess=Open-Source2026.07 | 72 | 84 | 60 | |
| DeepEyesAccess=Open-Source2026.07 | 71.25 | 83.75 | 58.75 | |
| MonetAccess=Open-Source2026.07 | 71 | 85.25 | 56.75 | |
| LVRAccess=Open-Source2026.07 | 70.62 | 83.5 | 57.75 | |
| Mull-TokensAccess=Open-Source2026.07 | 70.25 | 86.5 | 54 | |
| InternVL3-8BModel Category=Open-Source Models, Backbone=8B2026.06 | 70 | 78.8 | 61.3 | |
| Qwen2.5-VL-7BAccess=Open-Source, Model size=7B, Training=SFT + GRPO2026.07 | 70 | 83.25 | 56.75 | |
| Qwen2.5-VL-7BAccess=Open-Source, Model size=7B, Training=SFT2026.07 | 68.38 | 78.28 | 58.5 | |
| Qwen2.5-VL-7BModel Category=Open-Source Models, Backbone=7B2026.06 | 68 | 80.3 | 55.8 | |
| Qwen2.5-VL-7BAccess=Open-Source, Model size=7B2026.07 | 68 | 80.25 | 55.75 | |
| GPT-4oModel Category=Proprietary Model2026.06 | 65 | 66.8 | 63.3 | |
| GPT-4oAccess=Proprietary2026.07 | 59 | 70 | 48 |