GPT-5.6 is now available inside Figma Make, and it does something no previous model has done well inside a design tool: it takes a static design and produces a working, interactive prototype that actually looks like the original.
Figma announced the integration on July 9, and the results from early testing are significant enough that designers should pay attention. Not because another AI model joined the pile, but because this one specifically improves at the thing designers have been waiting for: turning a design vision into something clickable, responsive, and usable without starting from scratch in code.
What GPT-5.6 Does Inside Figma Make
Figma Make has been around since Config 2025. You type a prompt, it generates a build. The catch has always been quality. First passes looked rough. Layouts drifted from the original design. Interactions broke or never worked. You could get a prototype, but you spent as much time fixing it as you would have building it.
GPT-5.6 changes the equation. According to Figma’s own testing, three things are noticeably better:
Cleaner first passes. The model produces layouts with proper visual hierarchy, spacing, and styling on the first try. A prompt for a content-heavy e-commerce site came back with product descriptions, measurements, care guides, dropdowns that worked, an interactive photo library, and a clickable menu. All from one prompt.
Self-healing when things break. When a build hits an error, GPT-5.6 investigates and fixes it on its own instead of stopping dead. In one test, it identified a blank build, found the source of the problem, and resolved it without human intervention.
Faithful design-to-prototype conversion. This is the big one. Figma tested GPT-5.6 by feeding it an existing Figma Design file for a nature sound player app. The design included a multi-track timeline, playback controls, and a sound library with artwork. The first pass came back with the layout, visual hierarchy, spacing, proportions, and styling all matching the original. Play, pause, and skip controls worked. Audio files played back correctly.
That last point matters more than any benchmark. Designers have been asking for design to working prototype since Figma Make launched. This is the first model iteration where the output actually respects what the designer already created.
Three Model Tiers, One Design Tool
GPT-5.6 is not a single model. OpenAI released three tiers:
- Sol (gpt-5.6-sol): Flagship capability. The model Figma Make defaults to.
- Terra (gpt-5.6-terra): Balanced performance at lower cost.
- Luna (gpt-5.6-luna): Efficient, high-volume workloads.
Inside Figma Make, you pick your model from the model selector. Sol gives the best results. Terra and Luna trade quality for speed and token cost. For design work where visual fidelity matters, Sol is the one to use.
The model also introduces programmatic tool calling, which means it can write JavaScript to call tools, pass results between calls, and process intermediate outputs without waiting for you to prompt each step. Multi-agent support (in beta) lets it coordinate sub-agents in parallel for complex builds.
What You Can Actually Do With It
Here are the practical workflows that improve with GPT-5.6 inside Figma Make:
Design file to interactive prototype. Take an existing Figma Design file, prompt Make to build it, and get a working prototype that stays true to the original design. This used to require exporting to code or rebuilding interactions manually.
Rough sketch to polished prototype. Start with a napkin sketch or a text prompt. GPT-5.6 produces a usable first pass faster than previous models, which means you can explore more directions in a single session.
Responsive layouts from a single prompt. The e-commerce test showed that GPT-5.6 produces builds that adapt across screen sizes without additional prompting. This is still early, but the direction is promising.
Self-correcting builds. When something breaks, the model tries to fix it before handing it back. Less back-and-forth means you stay in your flow longer.
What to Look Out For
Token efficiency is better with GPT-5.6, but Make credits still matter. Figma published a separate guide on using credits efficiently, and it is worth reading before you start burning through builds.
The model has improved design judgment, but it is not perfect. First passes are closer to the vision than before, but they still need refinement. Think of it as a very good starting point, not a finished product.
Interactions work better but complex ones still need manual attention. The sound player test worked well, but that is one use case. Multi-step flows, conditional logic, and edge cases will likely need human review.
The programmatic tool calling and multi-agent features are new. They will get better, but right now they are best for bounded, tool-heavy workflows rather than open-ended creative exploration.
What This Unlocks
The gap between design and prototype has been shrinking for years. Figma Make with GPT-5.6 is the latest and most meaningful step. Designers who were already using Make will notice the jump immediately. Designers who wrote it off after early versions should give it another look.
The real unlock is speed. When a first pass is usable rather than rough, you can validate ideas faster, bring stakeholders in earlier, and spend less time translating between design and code. That is not a small thing. It changes the rhythm of how a design team works.
GPT-5.6 is available in Figma Make now. Open the model selector and pick it. Start with a design file you already have and see how close the first pass gets.
Want more like this? Join MOKU Club for free. Weekly resources, early access to new guides, and occasional templates you can actually use. Join below.
