SEAL-007 · Sealing-Groove Design · difficulty 3/5

ISO 3601-2 quad-ring face groove

sha256:1bcd01ef0a3c2200…

§1Prompt verbatim

Face-seal groove for an ISO 3601-2 quad-ring, nominal ID 25 mm, cross-section 3.0 mm. Static seal, low pressure (≤ 1 MPa). Standard squeeze 18-22 %, groove width 4.0 ± 0.05 mm.

§2Ground-truth spec

shells1

watertighttrue

manifoldtrue

acceptance ε±0.05 mm

featuresquad_ring_groove_25

§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✓

Human Baseline (Mech-E)

n=4 senior engineers

0.8189✓

GPT-5 → CadQuery

OpenAI + CadQuery 2.4

0.70310✓

OpenAI o4 (reasoning) → CadQuery

OpenAI + CadQuery 2.4

Claude Sonnet 4.6 → CadQuery

Anthropic + CadQuery 2.4

Qwen3 Coder → CadQuery

Alibaba + CadQuery 2.4

0.49711✗

Claude Opus 4.7 → CadQuery

Anthropic + CadQuery 2.4

0.47114✗

Claude Opus 4.7 → OpenSCAD

Anthropic + OpenSCAD 2024.06

0.4640✗

Claude Haiku 4.5 → CadQuery

Anthropic + CadQuery 2.4

0.45513✗

DeepSeek R1 (reasoning) → CadQuery

DeepSeek + CadQuery 2.4

0.44813✗

Gemini 2.5 Flash → CadQuery

Google + CadQuery 2.4

0.39716✗

Gemini 2.5 Pro → OpenSCAD

Google + OpenSCAD 2024.06

0.3910✗

DeepCAD

Wu et al. 2021 (research)

0.33116✗

CAD-Coder R1

CAD-Coder Labs (research)

0.30817✗

Llama 3.3 70B → OpenSCAD

Meta + OpenSCAD 2024.06

0.22625✗

GPT-5 Mini → OpenSCAD

OpenAI + OpenSCAD 2024.06

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.	Named-Dimension RMSE	Standards Compliance	P@1	p50	latency	cost
Human Baseline (Mech-E)	✓	0.979	0.110	0.943	1.000	—	783.3s	$5.934
GPT-5 → CadQuery	✓	0.962	0.237	0.642	0.000	—	42.1s	$0.216
OpenAI o4 (reasoning) → CadQuery	✓	0.954	0.188	0.781	0.000	—	126.2s	$1.260
Zoo Text-to-CAD	✓	0.943	0.188	0.726	0.000	—	8.0s	$0.163
Claude Sonnet 4.6 → CadQuery	✓	0.940	0.226	0.638	0.000	—	18.3s	$0.083
Adam (CADcrush)	✓	0.943	0.251	0.700	0.000	—	11.2s	$0.276
Qwen3 Coder → CadQuery	✓	0.925	0.327	0.525	0.000	—	14.6s	$0.027
Claude Opus 4.7 → CadQuery	✓	0.921	0.262	0.724	0.000	—	37.0s	$0.376
Claude Opus 4.7 → OpenSCAD	✓	0.919	0.206	0.595	0.000	—	27.7s	$0.268
Claude Haiku 4.5 → CadQuery	✓	0.919	0.290	0.462	0.000	—	9.4s	$0.020
DeepSeek R1 (reasoning) → CadQuery	✓	0.915	0.256	0.610	0.000	—	120.2s	$0.033
Gemini 2.5 Flash → CadQuery	×	0.908	0.304	0.516	0.000	—	9.9s	$0.021
Gemini 2.5 Pro → OpenSCAD	×	0.912	0.317	0.504	0.000	—	36.2s	$0.080
DeepCAD	×	0.904	0.366	0.274	0.000	—	3.9s	$0.023
CAD-Coder R1	×	0.900	0.285	0.527	0.000	—	5.2s	$0.006
Llama 3.3 70B → OpenSCAD	×	0.886	0.325	0.390	0.000	—	20.4s	$0.023
GPT-5 Mini → OpenSCAD	×	0.883	0.288	0.361	0.000	—	10.4s	$0.008
Trellis 3D	×	0.850	0.512	0.047	0.000	—	9.9s	$0.043
Spline AI	×	0.850	0.590	0.020	0.000	—	9.3s	$0.033
Hunyuan3D-2	×	0.850	0.527	0.048	0.000	—	28.2s	$0.083