Euclid-1
The world’s first purpose-built physical design model.
- Type
- Physical design model
- Base
- Post-trained on Inkling
- Input
- Written brief and multimodal references
- Output
- A feature-based part, built in a live CAD kernel
- Training
- RLFPC: Reinforcement Learning for Function and Physics Calibration
- Rewarded for
- Function, simulation performance and manufacturability
- Available in
- Enginuity Design Studio
Evaluation / ELCDB
A diverse dataset of complex parts, each graded on function, specification and construction quality with the reference part withheld. Every model gets the same brief, references and tools, in the Archimedes harness.
| Model | Functionality ↑ | Cost per task ↓ |
|---|---|---|
| Euclid-1 | 71.3 | $1.60 |
| GPT-6 Astra | 47.4 | $5.30 |
| Fable 5.1 | 38.6 | $2.80 |
| Inkling (base) | 13.1 | n/a |
