Spatial Reasoning on SpatialMQA
43.4AccuracyGUI-Cursor
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GUI-Cursorbackbone=Qwen2.5-VL-7B2025.09 | 43.4 | — | |
| Thinking with Grounding PointModel scale=4B, Grounding interface=Point, Grounding reward=w/o2026.06 | 39.13 | 67.19 | |
| Thinking with Grounding PointModel scale=4B, Grounding interface=Point, Grounding reward=w/2026.06 | 39.01 | 68.49 | |
| Gemma3-27B-ITModel scale=27B2026.06 | 38.99 | 54.28 | |
| Thinking with Grounding BoxModel scale=4B, Grounding interface=Box, Grounding reward=w/2026.06 | 38.68 | 68.49 | |
| Qwen2.5-VL-7B2025.09 | 38.1 | — | |
| GUI-G2-7B2025.09 | 38 | — | |
| Gemma3-12B-ITModel scale=12B2026.06 | 37.85 | 50 | |
| Thinking with Grounding BoxModel scale=4B, Grounding interface=Box, Grounding reward=w/o2026.06 | 37.64 | 67.66 | |
| LiquidEvaluation Protocol=Fine-tuned, Decoding Strategy=Greedy2025.06 | 28 | — | |
| L-FDMBackbone=Liquid, Evaluation Protocol=Forward-Dynamics Model, Decoding Strategy=Greedy2025.06 | 27.8 | — | |
| LiquidEvaluation Protocol=Zero-shot, Decoding Strategy=Greedy2025.06 | 27.7 | — | |
| C-FDMBackbone=Chameleon, Evaluation Protocol=Forward-Dynamics Model, Decoding Strategy=Greedy2025.06 | 27.2 | — | |
| ChameleonEvaluation Protocol=Zero-shot, Decoding Strategy=Greedy2025.06 | 26.1 | — | |
| ChameleonEvaluation Protocol=Fine-tuned, Decoding Strategy=Greedy2025.06 | 25.8 | — | |
| Gemma3-4B-ITModel scale=4B2026.06 | 25.35 | 36.43 | |
| Non-grounded ThinkingModel scale=4B, Grounding interface=None2026.06 | 14.17 | 27.88 |