GPT-5.6 Is Now in Figma Make — and First Drafts Just Got a Lot More Useful
Figma just dropped GPT-5.6 into Make, and early tests suggest it produces cleaner, more faithful prototypes on the very first pass. Here's what that actually means for designers who build.
Figma has added GPT-5.6 — OpenAI's newest model — to Figma Make, its AI-powered build tool. According to Figma's early tests, the upgrade meaningfully lifts both the quality and the speed of what Make produces before you've typed a single follow-up prompt.
That last part is the real headline. For designers who are starting to build their own apps, the hardest moment has always been the gap between a promising prompt and something that actually looks and behaves like the thing you imagined. GPT-5.6 narrows that gap.
What "first pass" quality actually means here
When Figma says "first pass," they mean the output you get before any refinement. That matters because every round of back-and-forth costs time, credits, and — more importantly — momentum.
In Figma's internal tests, they gave GPT-5.6 a minimal prompt to build a content-heavy e-commerce site with a bookshelf product catalog. The result came back with product descriptions, measurements, care guides, working dropdowns, an interactive photo library, and a clickable menu — all without additional prompting. The layout also adapted across screen sizes reliably, which is something that typically takes a few extra passes to get right.
This isn't about the AI doing everything perfectly. It's about starting closer to your vision, so the conversation you have with your team — or with yourself — is about refinement rather than rescue.
It stays faithful to your existing Figma designs
This is the part that should get designers most excited. GPT-5.6 is apparently quite good at reading a Figma Design file and building from it, not just from a text prompt.
In one of Figma's tests, they fed the model a design spec for a nature sound player app — complete with a multi-track timeline, playback controls, and a sound library with artwork. The first-pass build respected the layout, visual hierarchy, spacing, proportions, and styling. The interactions worked too: play, pause, and skip all functioned as expected, and audio files played back correctly.
For designers who already have polished mocks, this changes the workflow in a concrete way. Instead of describing your design to an AI and hoping it interprets your intent correctly, you can hand it the actual file. Think of it less as "generate something from scratch" and more as "animate my static comp."
It self-heals when something breaks
One quietly useful improvement: GPT-5.6 doesn't just stop when it hits an error. It investigates, identifies the problem, and fixes it on its own. Figma cites one case where the model caught the source of a blank build and resolved it without any input from the user.
For non-engineers, this matters a lot. A broken build used to mean either debugging unfamiliar code or starting over. Self-healing behaviour — where the model catches its own mistakes — keeps you in the flow instead of pulling you into troubleshooting territory you didn't sign up for.
How to try it right now
GPT-5.6 is live in Figma Make today. To switch to it:
- Open a project in Figma Make
- Find the model selector (top of the Make panel)
- Click on GPT-5.6 to activate it
Figma has also published a help center article on selecting and working with different models in Make — worth a quick skim if you want to understand when you might reach for one model over another.
One thing to keep in mind: Figma Make uses a credits system, and more capable models can use credits faster. Figma has a separate post on using Make credits more efficiently ("token efficiency" in AI terms just means getting more done per unit of compute). If you're building regularly, it's worth reading before you run a long session with the new model.
The honest takeaway
GPT-5.6 in Figma Make is a real upgrade, not a marketing version bump. Better first-pass quality, design-file fidelity, and self-healing behaviour are all things that directly reduce the friction of building for people who aren't developers.
What it doesn't change: you still need a clear idea of what you're building, and you'll still iterate. The model gets you closer to "good enough to share" faster — but it doesn't replace the design thinking that makes something worth sharing in the first place. That part is still yours.