High-Resolution Visual Reasoning on HR-Bench (4K/8K configurations)
81Score (4K)MAPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MAPObackbone=Ovis2.5-9B2026.04 | 81 | 78.6 | 79.8 | |
| Ovis2.5-9B + GRPOtraining=GRPO2026.04 | 78.5 | 77 | 77.8 | |
| Ovis2.5-9B + GSPOtraining=GSPO2026.04 | 78.3 | 76.5 | 77.4 | |
| Ovis2.5-9B + DAPOtraining=DAPO2026.04 | 78.1 | 76.1 | 77.1 | |
| DeepEyesv2results_source=original paper2026.04 | 77.9 | 73.8 | 75.9 | |
| Ovis2.5-9B + PPOtraining=PPO2026.04 | 77.4 | 76 | 76.7 | |
| Thymeresults_source=original paper2026.04 | 77 | 72 | 74.5 | |
| Mini-o3reproduced_by_authors=true2026.04 | 75.4 | 72.8 | 74.1 | |
| DeepeyesE2E=true, Param=7B2025.12 | 75 | 69.75 | 72.38 | |
| DeepEyesreproduced_by_authors=true2026.04 | 74.9 | 71.2 | 73 | |
| GPT-52026.04 | 74.6 | 72 | 73.3 | |
| Ovis2.5-9B + Coldstart SFTtraining=Supervised Fine-Tuning2026.04 | 74.6 | 71 | 72.8 | |
| Ovis2.5-9B2026.04 | 74 | 68.2 | 71.1 | |
| IMAgentE2E=true, Param=7B2025.12 | 73.5 | 71.5 | 72.5 | |
| Qwen2.5-VLParam=7B2025.12 | 70.62 | 66.75 | 68.69 | |
| ReGuLaR-7BModel Category=Latent reasoning LVLMs2026.05 | 70.5 | 61.88 | 66.19 | |
| LaserModel Category=Latent reasoning LVLMs2026.05 | 70.25 | 61.38 | 65.82 | |
| ZoomeyeE2E=false, Param=7B2025.12 | 69.6 | 69.3 | 69.45 | |
| DeepeyesModel Category=RL-based reasoning LVLMs2026.05 | 69.25 | 62 | 65.63 | |
| Pixel ReasonerE2E=false, Param=7B2025.12 | 68.75 | 63.25 | 66 | |
| Qwen2.5-VL-CoTParam=7B2025.12 | 68.5 | 62.25 | 65.38 | |
| Qwen2.5-VL-7BModel Category=General-purpose LVLMs2026.05 | 68 | 61.75 | 64.88 | |
| MonetModel Category=Latent reasoning LVLMs2026.05 | 67.37 | 60.88 | 64.13 | |
| Gemini 2.5 Pro2026.04 | 67.3 | 61.4 | 64.3 | |
| Vision-R1Model Category=RL-based reasoning LVLMs2026.05 | 66.63 | 60.25 | 63.44 | |
| LLaVA-OneVisionModel Category=General-purpose LVLMs2026.05 | 63.88 | 55.25 | 59.56 | |
| LLaVA-OneVisionParam=7B2025.12 | 63 | 59.8 | 61.4 | |
| GPT-4o2025.12 | 59 | 59 | 59 | |
| GPT-4oModel Category=General-purpose LVLMs2026.05 | 59 | 55.5 | 57.25 | |
| LVRModel Category=Latent reasoning LVLMs2026.05 | 51.13 | 50.13 | 50.63 |