Research2026-10-29
Figma is training a model to edit designs directly on its canvas. It uses supervised fine-tuning on 250,000 synthetic examples. The model now competes with top models on editing tasks. It also uses reinforcement learning to improve generalization. Evaluations check structure, visuals, and request fulfillment separately. Figma’s latest model scores 77.8% on its internal design editing tests. Sol 6.1 scores 76.9% on the same tests. The goal is faster, higher-quality edits for everyday design work.
What changed
Frontier models could not understand Figma syntax and required heavy system prompting.
What it unlocks
Making iterative design edits directly on the Figma canvas without manual intervention.
- 250,000 training examples
- 77.8% Figma design edits v3 score
- 76.9% Sol 6.1 score
What you need to act on it
- Figma native file format
- design system assets
- supervised fine-tuning and reinforcement learning
Sources