DFMMOLD-005 · DFM · Injection Mould · difficulty 5/5

Telephone handset shell

sha256:0ec1bd0a02fe1011…

§1Prompt verbatim

Lower handset shell, 180 × 60 × 22 mm. Parting plane is the largest XY silhouette. Min draft 1.5° on every wall, uniform 1.8 mm wall ±10 %, four self-tapping bosses Ø 5 OD / Ø 2.5 core, mic and speaker mesh openings 12 × 8 mm slot arrays.

§2Ground-truth spec

shells1

watertighttrue

manifoldtrue

acceptance ε±0.1 mm

featuresdraft_1.5deg_min, wall_1.8mm_uniform, boss_x4, slot_array

§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

0.63712✓

OpenAI o4 (reasoning) → CadQuery

OpenAI + CadQuery 2.4

0.56810✗

GPT-5 → CadQuery

OpenAI + CadQuery 2.4

Claude Sonnet 4.6 → CadQuery

Anthropic + CadQuery 2.4

0.48012✗

Claude Opus 4.7 → CadQuery

Anthropic + CadQuery 2.4

0.44916✗

DeepSeek R1 (reasoning) → CadQuery

DeepSeek + CadQuery 2.4

0.41915✗

Claude Opus 4.7 → OpenSCAD

Anthropic + OpenSCAD 2024.06

0.4060✗

CAD-Coder R1

CAD-Coder Labs (research)

Gemini 2.5 Pro → OpenSCAD

Google + OpenSCAD 2024.06

0.3570✗

Claude Haiku 4.5 → CadQuery

Anthropic + CadQuery 2.4

0.35616✗

DeepCAD

Wu et al. 2021 (research)

0.30318✗

Qwen3 Coder → CadQuery

Alibaba + CadQuery 2.4

0.25022✗

Gemini 2.5 Flash → CadQuery

Google + CadQuery 2.4

0.24522✗

Llama 3.3 70B → OpenSCAD

Meta + OpenSCAD 2024.06

0.17428✗

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.	Draft-Angle Compliance	Min-Wall Compliance	Wall-Thickness Uniformity	p50	latency	cost
Human Baseline (Mech-E)	✓	0.952	0.881	0.789	0.898	—	543.0s	$5.057
OpenAI o4 (reasoning) → CadQuery	✓	0.931	0.606	0.641	0.630	—	109.7s	$1.142
GPT-5 → CadQuery	✓	0.934	0.613	0.598	0.606	—	46.4s	$0.206
Zoo Text-to-CAD	✓	0.929	0.570	0.630	0.731	—	6.1s	$0.179
Claude Sonnet 4.6 → CadQuery	✓	0.922	0.623	0.623	0.599	—	22.0s	$0.070
Claude Opus 4.7 → CadQuery	✓	0.915	0.650	0.678	0.711	—	31.0s	$0.282
DeepSeek R1 (reasoning) → CadQuery	✓	0.918	0.545	0.508	0.546	—	104.6s	$0.041
Claude Opus 4.7 → OpenSCAD	×	0.913	0.500	0.569	0.577	—	24.7s	$0.282
CAD-Coder R1	×	0.904	0.401	0.510	0.542	—	4.3s	$0.005
Adam (CADcrush)	×	0.906	0.675	0.575	0.683	—	10.8s	$0.319
Gemini 2.5 Pro → OpenSCAD	×	0.902	0.417	0.515	0.511	—	26.3s	$0.088
Claude Haiku 4.5 → CadQuery	×	0.900	0.373	0.424	0.417	—	8.8s	$0.017
DeepCAD	×	0.894	0.382	0.329	0.403	—	5.4s	$0.016
Qwen3 Coder → CadQuery	×	0.889	0.442	0.506	0.481	—	14.4s	$0.029
Gemini 2.5 Flash → CadQuery	×	0.887	0.536	0.555	0.542	—	10.7s	$0.017
Llama 3.3 70B → OpenSCAD	×	0.877	0.317	0.457	0.361	—	25.7s	$0.018
GPT-5 Mini → OpenSCAD	×	0.868	0.382	0.436	0.428	—	15.7s	$0.010
Trellis 3D	×	0.850	0.100	0.230	0.129	—	14.6s	$0.041
Spline AI	×	0.850	0.047	0.100	0.064	—	9.7s	$0.037
Hunyuan3D-2	×	0.850	0.092	0.243	0.133	—	24.5s	$0.071