STD-005 · Standards Compliance · difficulty 3/5

ISO 8734 dowel pin Ø6 m6 × 30

sha256:00b41ef02ac9d301…

§1Prompt verbatim

Cylindrical dowel pin per ISO 8734 type A: Ø 6 m6 (+0.012 / +0.004), length 30 mm, both ends spherically radiused R 0.6 mm. Surface finish Ra ≤ 0.4 µm.

§2Ground-truth spec

shells1

watertighttrue

manifoldtrue

acceptance ε±0.005 mm

§3Reference render

canonical reference · drag to orbit, scroll to zoom

Visualisation is rebuilt in-browser from the canonical parametric description. Scoring is performed against the held-out reference STEP file (sha-256 fingerprint above).

§4Per-agent renders

reference + 10 agent outputs · scored against the held-out STEP

vol IoU · BREP · manifold

canonical reference

REFERENCE

canonical · ground truth

1.000100✓

OpenAI o4 (reasoning) → CadQuery

OpenAI + CadQuery 2.4

0.86610✓

Human Baseline (Mech-E)

Gemini 2.5 Flash → CadQuery

Google + CadQuery 2.4

0.62510✗

Claude Sonnet 4.6 → CadQuery

Anthropic + CadQuery 2.4

0.6129✗

GPT-5 → CadQuery

OpenAI + CadQuery 2.4

0.61011✗

Claude Opus 4.7 → CadQuery

Anthropic + CadQuery 2.4

Claude Haiku 4.5 → CadQuery

Anthropic + CadQuery 2.4

0.49413✗

Claude Opus 4.7 → OpenSCAD

Anthropic + OpenSCAD 2024.06

0.4780✗

Llama 3.3 70B → OpenSCAD

Meta + OpenSCAD 2024.06

0.41815✗

Qwen3 Coder → CadQuery

Alibaba + CadQuery 2.4

0.40314✗

CAD-Coder R1

CAD-Coder Labs (research)

0.39116✗

DeepSeek R1 (reasoning) → CadQuery

DeepSeek + CadQuery 2.4

0.36815✗

Gemini 2.5 Pro → OpenSCAD

Google + OpenSCAD 2024.06

0.3240✗

DeepCAD

Wu et al. 2021 (research)

0.23923✗

GPT-5 Mini → OpenSCAD

OpenAI + OpenSCAD 2024.06

no manifold solid produced

Each tile is rebuilt from the canonical parametric description and degraded to match the agent's scored profile (tessellation, non-manifold face removal, dimension scale jitter, missing features). Image-only diffusion models render visually plausible meshes but score in the single digits on BREP fidelity — the geometry is not a manifold solid even when the render reads clean.

§5Per-agent metrics

ranked by Vol IoU · same data as the leaderboard, restricted to this task

Agent	Watert.	Manif.	GD&T Compliance	Standards Compliance	P@1	p50	latency	cost
OpenAI o4 (reasoning) → CadQuery	✓	0.973	0.623	0.771	1.000	—	106.0s	$0.880
Human Baseline (Mech-E)	✓	0.961	0.833	0.900	1.000	—	628.0s	$5.331
Zoo Text-to-CAD	✓	0.950	0.722	0.744	0.000	—	4.4s	$0.152
Gemini 2.5 Flash → CadQuery	✓	0.938	0.463	0.583	0.000	—	12.7s	$0.024
Claude Sonnet 4.6 → CadQuery	✓	0.938	0.588	0.680	0.000	—	13.3s	$0.074
GPT-5 → CadQuery	✓	0.948	0.606	0.736	0.000	—	45.6s	$0.224
Claude Opus 4.7 → CadQuery	✓	0.936	0.622	0.735	0.000	—	47.9s	$0.315
Adam (CADcrush)	✓	0.938	0.692	0.716	0.000	—	7.5s	$0.281
Claude Haiku 4.5 → CadQuery	✓	0.922	0.394	0.469	0.000	—	10.3s	$0.020
Claude Opus 4.7 → OpenSCAD	✓	0.927	0.430	0.557	0.000	—	24.2s	$0.309
Llama 3.3 70B → OpenSCAD	×	0.913	0.283	0.386	0.000	—	15.4s	$0.021
Qwen3 Coder → CadQuery	×	0.911	0.424	0.520	0.000	—	13.9s	$0.033
CAD-Coder R1	×	0.913	0.468	0.491	0.000	—	5.2s	$0.005
DeepSeek R1 (reasoning) → CadQuery	×	0.902	0.526	0.533	0.000	—	104.6s	$0.035
Gemini 2.5 Pro → OpenSCAD	×	0.895	0.443	0.479	0.000	—	22.1s	$0.095
DeepCAD	×	0.885	0.373	0.281	0.000	—	3.5s	$0.021
GPT-5 Mini → OpenSCAD	×	0.881	0.269	0.379	0.000	—	10.6s	$0.010
Trellis 3D	×	0.868	0.059	0.047	0.000	—	13.4s	$0.041
Spline AI kernel error: BRepCheck_NotClosed	×	0.000	—	—	0.000	—	8.5s	$0.034
Hunyuan3D-2	×	0.850	0.056	0.043	0.000	—	38.8s	$0.064