Benchmarks for AI Models and Agents on CAD Tasks
gNucleus-CAD-Bench is a comprehensive collection of benchmarks to benchmark CAD models and AI agents on CAD design and 3D modeling tasks.
Hugging Face
Leaderboard
View full leaderboard →| Rank | Name | Geometry | Parametric | Spec Consistency | Overall |
|---|---|---|---|---|---|
1 | Claude Opus 4.7 | 89.2 | 86.5 | 87.0 | 87.6 |
2 | GPT-5.3-Codex | 86.8 | 84.2 | 83.5 | 85.0 |
3 | Claude Opus 4.5 | 82.5 | 80.4 | 79.8 | 80.9 |
4 | Claude Opus 4.6 | 82.1 | 80.5 | 79.6 | 80.8 |
5 | Gemini 3.1 Pro | 81.9 | 80.8 | 79.0 | 80.6 |
6 | MiniMax M2.5 | 81.0 | 79.8 | 79.3 | 80.2 |
7 | GPT-5.2 | 80.7 | 80.1 | 78.5 | 80.0 |
8 | Qwen3.6 Plus | 80.0 | 78.4 | 77.5 | 78.8 |
9 | GLM-5 | 78.8 | 77.6 | 76.9 | 77.8 |
10 | Muse Spark | 78.5 | 77.0 | 76.5 | 77.4 |
Scores last refreshed May 12, 2024. Higher is better.
Benchmarks Tasks
3D Parametric Part Generation
Generate editable, parametric 3D CAD part models from natural-language prompts and reference inputs.
Assembly Generation
Generate multi-part assemblies with proper mates, constraints, and component hierarchy.
Complex CAD workflow
Multi-step CAD workflows that generate, iteratively edit, and verify designs until the model meets the target spec.
Evaluation Methods
Geometry Accuracy
Measures how closely the generated geometry matches the reference part.
Constraint & Assembly Correctness
Checks constraint satisfaction, mating validity, and assembly stability.
Parametric Correctness
Verifies CAD model consistency with the spec parameters.
Topology / Structure
Evaluates topological validity and part structure correctness.
Agent Workflow Success
Assesses task completion rate and workflow correctness.
Efficiency
Measures token usage, execution time, and resource efficiency.
Latest Updates
May 12, 2024
Benchmark v1.2 release
Added 12 new assembly tasks and refreshed leaderboard.
May 5, 2024
New evaluation tool: TopoCheck
Improved topology evaluation now supports curved surfaces.
Apr 14, 2024
Open contributions
External submissions are now accepted via the GitHub repo.