CAD-Bench
← back
CAL-003 · Confidence Calibration · difficulty 5/5

Confidence-calibrated planetary carrier

sha256:2b97cc4d1ef0aa55…

§1Prompt verbatim

Build the planetary carrier of MECH-027 and report a self-assessed pre-generation confidence ∈ [0,1] in your output's correctness against the spec.

§2Ground-truth spec

shells1
watertighttrue
manifoldtrue
acceptance ε±0.05 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)
Human Baseline (Mech-E)
n=4 senior engineers
0.5998✗
OpenAI o4 (reasoning) → CadQuery
OpenAI o4 (reasoning) → CadQuery
OpenAI + CadQuery 2.4
0.50314✗
DeepSeek R1 (reasoning) → CadQuery
DeepSeek R1 (reasoning) → CadQuery
DeepSeek + CadQuery 2.4
0.40615✗
Claude Opus 4.7 → OpenSCAD
Claude Opus 4.7 → OpenSCAD
Anthropic + OpenSCAD 2024.06
0.3990✗
Claude Sonnet 4.6 → CadQuery
Claude Sonnet 4.6 → CadQuery
Anthropic + CadQuery 2.4
0.38614✗
Claude Opus 4.7 → CadQuery
Claude Opus 4.7 → CadQuery
Anthropic + CadQuery 2.4
0.37115✗
Qwen3 Coder → CadQuery
Qwen3 Coder → CadQuery
Alibaba + CadQuery 2.4
0.36618✗
Adam (CADcrush)
Adam (CADcrush)
CADcrush
0.36416✗
Claude Haiku 4.5 → CadQuery
Claude Haiku 4.5 → CadQuery
Anthropic + CadQuery 2.4
0.35513✗
GPT-5 → CadQuery
GPT-5 → CadQuery
OpenAI + CadQuery 2.4
0.34015✗
Zoo Text-to-CAD
Zoo Text-to-CAD
Zoo (KittyCAD)
0.32120✗
GPT-5 Mini → OpenSCAD
GPT-5 Mini → OpenSCAD
OpenAI + OpenSCAD 2024.06
0.30216✗
CAD-Coder R1
CAD-Coder R1
CAD-Coder Labs (research)
0.29918✗
Gemini 2.5 Pro → OpenSCAD
Gemini 2.5 Pro → OpenSCAD
Google + OpenSCAD 2024.06
0.2790✗
Llama 3.3 70B → OpenSCAD
Llama 3.3 70B → OpenSCAD
Meta + OpenSCAD 2024.06
0.23822✗
DeepCAD
DeepCAD
Wu et al. 2021 (research)
0.07464✗
Trellis 3D
Trellis 3D
Microsoft Research
0.0250✗
Spline AI
Spline AI
Spline.design
0.0000✗
no manifold solid produced
Gemini 2.5 Flash → CadQuery
Gemini 2.5 Flash → CadQuery
Google + CadQuery 2.4
—59✗
no manifold solid produced
Hunyuan3D-2
Hunyuan3D-2
Tencent
—5✗

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
AgentWatert.Manif.Calibration (Brier)P@1p50latencycost
Human Baseline (Mech-E)✓0.9420.0340.000—633.7s$7.134
OpenAI o4 (reasoning) → CadQuery✓0.9260.0700.000—131.8s$0.967
DeepSeek R1 (reasoning) → CadQuery✓0.9130.1500.000—78.6s$0.034
Claude Opus 4.7 → OpenSCAD×0.9120.0970.000—27.8s$0.290
Claude Sonnet 4.6 → CadQuery×0.9060.1240.000—18.3s$0.073
Claude Opus 4.7 → CadQuery×0.9030.1150.000—33.0s$0.387
Qwen3 Coder → CadQuery×0.9020.1690.000—18.6s$0.034
Adam (CADcrush)×0.9020.1610.000—9.5s$0.268
Claude Haiku 4.5 → CadQuery×0.9080.1690.000—8.3s$0.022
GPT-5 → CadQuery×0.9000.1100.000—36.6s$0.185
Zoo Text-to-CAD×0.8960.1060.000—5.8s$0.158
GPT-5 Mini → OpenSCAD×0.8970.1560.000—10.1s$0.011
CAD-Coder R1×0.8990.1540.000—6.1s$0.005
Gemini 2.5 Pro → OpenSCAD×0.8900.1140.000—35.8s$0.083
Llama 3.3 70B → OpenSCAD×0.8870.1290.000—17.1s$0.020
DeepCAD×0.8610.1690.000—5.2s$0.022
Trellis 3D×0.8540.2050.000—9.9s$0.049
Spline AI×0.8500.1890.000—9.8s$0.039
Gemini 2.5 Flash → CadQuery
kernel error: BRepCheck_NotClosed
×0.000—0.000—14.5s$0.021
Hunyuan3D-2
kernel error: BRepCheck_NotClosed
×0.000—0.000—30.0s$0.074