Spatial Reasoning on HR-Bench 8K
73.3HR-8K Average AccuracyMini-o3
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Mini-o3SFT=true, RL=true2026.04 | 73.3 | — | — | |
| ToolsRLSFT=false, RL=true2026.04 | 73.2 | 88.1 | 58.3 | |
| DeepEyesSFT=false, RL=true2026.04 | 72.7 | 86.8 | 58.5 | |
| ZoomEyeSFT=false, RL=false2026.04 | 69.3 | 88.5 | 50 | |
| Pixel-ReasonerSFT=true, RL=true2026.04 | 66.9 | — | — | |
| Qwen2.5-VLSFT=false, RL=false2026.04 | 65.3 | 78.8 | 51.8 | |
| Point-RFTSFT=true, RL=true2026.04 | — | 36.2 | — |