Brick Selection on BC-Bench Best Object Performance
93.1Selection AccuracyGPT-5.4
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-5.4Evaluation Protocol=Zero-shot2026.06 | 93.1 | |
| Brick-ComposerBackbone=Qwen-3-8B-VL, Evaluation Protocol=Integrated learning2026.06 | 90.65 | |
| Qwen-3.5-VL-27BEvaluation Protocol=Zero-shot2026.06 | 75.86 | |
| Designer SupervisionBackbone=Qwen-3-8B-VL, Evaluation Protocol=Human supervision2026.06 | 73.24 | |
| Qwen-3-VL-8BEvaluation Protocol=Zero-shot2026.06 | 55.17 | |
| Brick-ComposerBackbone=Gemma-3-12B, Evaluation Protocol=Integrated learning2026.06 | 52.87 | |
| InternVL-3.5-8BEvaluation Protocol=Zero-shot2026.06 | 31.25 | |
| Designer SupervisionBackbone=Gemma-3-12B, Evaluation Protocol=Human supervision2026.06 | 26.67 | |
| Gemma-3-12BEvaluation Protocol=Zero-shot2026.06 | 17.24 |