Single-image spatial reasoning on SPBench 2025
68.4AccuracySpatialClaw
Evaluation Results
| Method | Links | |
|---|---|---|
| SpatialClawModel=Gemma4-31B2026.06 | 68.4 | |
| SpatialClawModel=Qwen3.6-27B2026.06 | 64.8 | |
| SpatialClawModel=Qwen3.5-122B-A10B2026.06 | 63.6 | |
| SpatialClawModel=Qwen3.5-397B-A17B2026.06 | 63.3 | |
| SpatialClawModel=Qwen3.6-35B-A3B2026.06 | 63 | |
| SpatialClawModel=Gemma4-26B-A4B2026.06 | 61.9 | |
| No-toolModel=Qwen3.6-27B2026.06 | 58 | |
| No-toolModel=Gemma4-31B2026.06 | 55.1 | |
| No-toolModel=Qwen3.5-397B-A17B2026.06 | 54 | |
| No-toolModel=Qwen3.6-35B-A3B2026.06 | 52.4 | |
| No-toolModel=Qwen3.5-122B-A10B2026.06 | 52.2 | |
| No-toolModel=Gemma4-26B-A4B2026.06 | 45.1 |