Brick Selection on BC-Bench Full (test)
68.21Selection AccuracyBrick-Composer
Evaluation Results
| Method | Links | |
|---|---|---|
| Brick-ComposerBackbone=Qwen-3-8B-VL, Evaluation Protocol=Integrated learning2026.06 | 68.21 | |
| Designer SupervisionBackbone=Qwen-3-8B-VL, Evaluation Protocol=Human supervision2026.06 | 48.29 | |
| GPT-5.4Evaluation Protocol=Zero-shot2026.06 | 43.88 | |
| Qwen-3.5-VL-27BEvaluation Protocol=Zero-shot2026.06 | 37.44 | |
| Qwen-3-VL-8BEvaluation Protocol=Zero-shot2026.06 | 22.76 | |
| Brick-ComposerBackbone=Gemma-3-12B, Evaluation Protocol=Integrated learning2026.06 | 17.95 | |
| Designer SupervisionBackbone=Gemma-3-12B, Evaluation Protocol=Human supervision2026.06 | 15.59 | |
| InternVL-3.5-8BEvaluation Protocol=Zero-shot2026.06 | 13.24 | |
| Gemma-3-12BEvaluation Protocol=Zero-shot2026.06 | 4.35 |