Joint Assembly Evaluation on BC-Bench Full (test)
14.27Step-Wise SRBrick-Composer
Evaluation Results
| Method | Links | |
|---|---|---|
| Brick-ComposerBackbone=Qwen-3-8B-VL, Evaluation Protocol=Integrated learning2026.06 | 14.27 | |
| World Feedback (L)Backbone=Qwen-3-8B-VL, Evaluation Protocol=Simulator-generated feedback2026.06 | 6.24 | |
| Designer SupervisionBackbone=Qwen-3-8B-VL, Evaluation Protocol=Human supervision2026.06 | 5.4 | |
| Brick-ComposerBackbone=Gemma-3-12B, Evaluation Protocol=Integrated learning2026.06 | 4.35 | |
| World Feedback (L)Backbone=Gemma-3-12B, Evaluation Protocol=Simulator-generated feedback2026.06 | 1.99 | |
| Designer SupervisionBackbone=Gemma-3-12B, Evaluation Protocol=Human supervision2026.06 | 0.72 | |
| Qwen-3-VL-8BEvaluation Protocol=Zero-shot2026.06 | 0.36 | |
| World Feedback (P)Backbone=Qwen-3-8B-VL, Evaluation Protocol=Simulator feedback2026.06 | 0.27 | |
| Qwen-3.5-VL-27BEvaluation Protocol=Zero-shot2026.06 | 0.18 | |
| GPT-5.4Evaluation Protocol=Zero-shot2026.06 | 0.18 | |
| Gemma-3-12BEvaluation Protocol=Zero-shot2026.06 | 0.09 | |
| InternVL-3.5-8BEvaluation Protocol=Zero-shot2026.06 | 0 | |
| World Feedback (P)Backbone=Gemma-3-12B, Evaluation Protocol=Simulator feedback2026.06 | 0 |