OmniMech - CAD Program Pass Rate (2D Drawing): leaderboard
OmniMech (2026) asks vision-language models to turn an industrial 2D engineering drawing with dimensions and tolerances into an executable parametric CAD program. This board is the percentage of 5,000 sampled drawings for which the model returns a complete program that builds a valid, non-empty 3D model, from the drawing alone; higher is better.
Metric: Pass rate (%; input: the 2D engineering drawing only; 5,000 sampled industrial 2D orthographic drawings with dimension and tolerance annotations, zero-shot; a trial passes only when the model returns a complete response whose extracted CAD program builds a valid, non-empty 3D model). Source: arxiv.org. Saturation forecast: Estimated already saturated. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.8 | 92.7 | #56 |
| 2 | GPT-5.5 | 92.6 | #49 |
| 3 | Gemini 3.1 Pro (Preview) | 81.1 | #94 |
| 4 | Gemini 3.5 Flash | 70.6 | #101 |
| 5 | Qwen 3.7 Plus | 66.8 | #207 |
| 6 | Llama 4 Maverick | 61 | #1099 |
| 7 | GPT-5.4 Mini | 60 | #389 |
| 8 | Qwen 3.5 9B | 1.3 | #803 |
| 9 | Ministral 3 14B | 0.4 | #1845 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=omnimech-cad-program-pass-rate-2d-drawing · How It Works · Data refreshed daily, snapshot 2026-11-13.