Skip to content

Dataset · gnucleus-ai

FreeCAD 3D modeling benchmark runs

gnucleus-ai/cad-gen-freecad-bench-v2

A table of results from 1,000 runs of AI agents on 100 parametric CAD generation tasks from text.

What's inside

The archive contains tabular and text data in parquet format (data/cad-bench-v2.parquet) with evaluation results of agents on the Parametric CAD Bench v2 benchmark. Inside are 1,000 rows—one attempt for each of the 100 tasks across 10 public runs from the gnucleus-ai/cad-bench@v2 leaderboard. Each entry contains a continuous reward score for generating editable FreeCAD models from text descriptions.

What you can do with it

  • Analyzing errors and success rates of AI agents in CAD model generation
  • Studying quality metrics in text-to-3D modeling tasks
  • Comparing model runs on the Parametric CAD Bench v2 benchmark
Size
1K–10K rows
Downloads this month
26.1K
Likes
2

How to load it

from datasets import load_dataset

ds = load_dataset("gnucleus-ai/cad-gen-freecad-bench-v2")