ASM-001 · Assembly & Mating · difficulty 4/5

Threaded coupling M16×1.5 (male+female pair)

sha256:33ce1d0bc09f01ea…

§1Prompt verbatim

Two-piece threaded coupling. Male: Ø 16 × 30 mm long with M16×1.5 6g external thread, hex Ø 22 across-flats × 8 mm tall. Female: Ø 24 × 25 mm with M16×1.5 6H internal thread depth 22 mm, hex Ø 22. The pair must thread together a full 18 mm with no interference.

§2Ground-truth spec

shells1

watertighttrue

manifoldtrue

acceptance ε±0.02 mm

clearance[0, 0.4] 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✓

Human Baseline (Mech-E)

OpenAI o4 (reasoning) → CadQuery

OpenAI + CadQuery 2.4

Qwen3 Coder → CadQuery

Alibaba + CadQuery 2.4

0.48712✗

DeepSeek R1 (reasoning) → CadQuery

DeepSeek + CadQuery 2.4

0.47613✗

Claude Opus 4.7 → CadQuery

Anthropic + CadQuery 2.4

0.46412✗

Claude Opus 4.7 → OpenSCAD

Anthropic + OpenSCAD 2024.06

0.4500✗

Claude Sonnet 4.6 → CadQuery

Anthropic + CadQuery 2.4

0.44513✗

GPT-5 → CadQuery

OpenAI + CadQuery 2.4

0.43112✗

Gemini 2.5 Pro → OpenSCAD

Google + OpenSCAD 2024.06

0.3860✗

Gemini 2.5 Flash → CadQuery

Google + CadQuery 2.4

0.37114✗

CAD-Coder R1

CAD-Coder Labs (research)

0.32814✗

GPT-5 Mini → OpenSCAD

OpenAI + OpenSCAD 2024.06

0.31117✗

DeepCAD

Wu et al. 2021 (research)

0.26317✗

Llama 3.3 70B → OpenSCAD

Meta + OpenSCAD 2024.06

0.23422✗

Claude Haiku 4.5 → CadQuery

Anthropic + CadQuery 2.4

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.	FeatRec	Mating Clearance	Fit-Class Compliance	P@1	p50	latency	cost
Human Baseline (Mech-E)	✓	0.965	0.841	0.808	0.764	1.000	—	621.8s	$5.263
Zoo Text-to-CAD	✓	0.953	0.755	0.693	0.495	0.000	—	6.3s	$0.160
OpenAI o4 (reasoning) → CadQuery	✓	0.943	0.795	0.631	0.453	0.000	—	102.2s	$0.978
Adam (CADcrush)	✓	0.939	0.646	0.644	0.508	0.000	—	10.2s	$0.226
Qwen3 Coder → CadQuery	✓	0.927	0.584	0.494	0.215	0.000	—	19.2s	$0.034
DeepSeek R1 (reasoning) → CadQuery	✓	0.928	0.626	0.562	0.309	0.000	—	116.8s	$0.041
Claude Opus 4.7 → CadQuery	✓	0.915	0.776	0.636	0.458	0.000	—	46.7s	$0.293
Claude Opus 4.7 → OpenSCAD	✓	0.915	0.594	0.479	0.231	0.000	—	22.8s	$0.313
Claude Sonnet 4.6 → CadQuery	✓	0.919	0.731	0.620	0.388	0.000	—	13.5s	$0.077
GPT-5 → CadQuery	✓	0.919	0.652	0.543	0.395	0.000	—	36.4s	$0.232
Gemini 2.5 Pro → OpenSCAD	×	0.906	0.505	0.478	0.216	0.000	—	34.6s	$0.087
Gemini 2.5 Flash → CadQuery	×	0.904	0.543	0.517	0.243	0.000	—	11.9s	$0.020
CAD-Coder R1	×	0.899	0.667	0.472	0.255	0.000	—	5.6s	$0.006
GPT-5 Mini → OpenSCAD	×	0.896	0.396	0.343	0.116	0.000	—	10.8s	$0.011
DeepCAD	×	0.893	0.475	0.395	0.167	0.000	—	5.4s	$0.021
Llama 3.3 70B → OpenSCAD	×	0.887	0.430	0.399	0.112	0.000	—	15.8s	$0.022
Claude Haiku 4.5 → CadQuery	×	0.885	0.497	0.363	0.176	0.000	—	10.0s	$0.024
Trellis 3D	×	0.850	0.218	0.095	0.005	0.000	—	15.0s	$0.043
Spline AI	×	0.850	0.090	0.050	0.001	0.000	—	9.0s	$0.042
Hunyuan3D-2	×	0.850	0.210	0.082	0.005	0.000	—	26.1s	$0.073