Two-piece spatial reasoning on Tangram
34Pos IoUGemini-2.5-pro
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Gemini-2.5-pro2026.02 | 34 | 39.7 | 34 | |
| Claude-Sonnet-42026.02 | 31.8 | 39.4 | 23.5 | |
| GPT-4o mini2026.02 | 27.6 | 39.4 | 27.8 | |
| Qwen-72B2026.02 | 25.3 | 49.5 | 24.8 | |
| LLaMA Maverick2026.02 | 22 | 42.7 | 37.1 | |
| Qwen-3B2026.02 | 19.2 | 31.7 | 21.4 |