Single-image spatial reasoning on SPBench
68.4AccuracySpatialClaw
Evaluation Results
| Method | Links | |
|---|---|---|
| SpatialClawBackbone=Gemma4-31B2026.06 | 68.4 | |
| pySpatialBackbone=Gemma4-31B2026.06 | 53.5 | |
| SpaceTools ToolshedBackbone=Gemma4-31B2026.06 | 45.1 | |
| VADARBackbone=Gemma4-31B2026.06 | 41.6 |