🛠️ MUSE Leaderboard

Multimodal Understanding of Spatial Engineering — LLM-generated CadQuery from natural-language design specs

Stage 1 · Code Check
Sandbox Success
Did the generated CadQuery script execute without error?
Stage 2 · Geometry Check
Geometric Validity
All four OCCT checks pass: watertight, manifold, self-intersection free, overlap free.
Stage 3 · Design Intent
VLM Judge (Gemini-3.1-Pro)
Three pillars: Functionality · Manufacturability · Assemblability. Failed upstream ⇒ all Stage 3 forced to 0.
Loading…