Spatial Reasoning on HR-Bench 4K
77.5HR-4K Average ScoreMini-o3
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Mini-o3SFT=true, RL=true2026.04 | 77.5 | — | — | |
| Simple o3SFT=true, RL=false2026.04 | 76.2 | — | — | |
| ToolsRLSFT=false, RL=true2026.04 | 75.9 | 91.2 | 60.6 | |
| DeepEyesSFT=false, RL=true2026.04 | 75.2 | 91.3 | 59 | |
| Pixel-ReasonerSFT=true, RL=true2026.04 | 74 | — | — | |
| Qwen2.5-VLSFT=false, RL=false2026.04 | 70.4 | 83.8 | 56.9 | |
| ZoomEyeSFT=false, RL=false2026.04 | 69.7 | 84.3 | 55 |