BrickBench: Evaluating Agentic Brick Design
The benchmark tests whether coding agents can design LEGO assemblies that match a prompt and can actually be built.
BrickBench scores validity, alignment, and design across three settings with different scale and part constraints. The authors also provide BrickAgent, an environment for agents to construct, inspect, and validate designs. Their finding: leading agents handle many verifiable physical and semantic requirements, but still trail human designs. ArXiv · AI/CL/LG's note
BrickBench scores validity, alignment, and design across three settings with different scale and part constraints. The authors also provide BrickAgent, an environment for agents to construct, inspect, and validate designs. Their finding: leading agents handle many verifiable physical and semantic requirements, but still trail human designs. ArXiv · AI/CL/LG's note
score 5