Both tools generate a screen from a paragraph. That is the demo. The real split is where the file lives after the demo. Eigent’s 2026 comparison puts it without the keynote: Claude Design is strongest when you want visuals grounded in a design system and a codebase. Figma Make is strongest when you already work in Figma and you want to poke the result on the canvas.
I am giving the win to Figma Make for teams that already have a library. That is most of the readers who will actually pay for this. Claude Design wins the first afternoon for a PM who hates opening Figma. Those are different buyers. This site still needs a winner field.
What each product is
Claude Design, in research preview, sits inside Anthropic’s existing subscriptions. You talk, it produces prototypes, slides, one-pagers. Eigent says it can take your real design files and code as gravity. The output is HTML-ish, not a .fig with variants.
Figma Make is prompt-to-prototype inside Figma. You generate, then you point-and-edit. Figma’s own blog on Claude Code to Figma is a sibling feature: capture a live UI from production, staging, or localhost and land it as editable frames. Make stays in the file. Code-to-canvas is how developers join the same file without redrawing a dashboard by hand.
Yiannis Antoniou’s Substack is the most honest user write-up I found this round. First generation quality impressed him. Design-system import felt like it understood a system, not just a hex dump. The Claude Code handoff felt like a path. Then the billboard: token cost makes real iteration painful on Pro and tight even on Max. There is no native Figma export. You dump HTML and run it through a third party. Claude Code, by contrast, already has two-way Figma MCP.
If you remember nothing else: Design generates. Make inhabits. Code-to-Figma imports. Those are three verbs. Marketing will mash them.
Canva versus Adobe Express is a different fight, the one we already ran in 2026. This is not that. Affinity versus Photoshop is still pixels and licenses. Claude versus ChatGPT as chatbots is a third article. Do not shop with the wrong pair.
Pricing is the first real difference
Claude Design, per Eigent, is bundled. Claude Pro is $20 a month. Max, Team, and Enterprise sit above that. No separate per-project fee in preview. You already pay for Claude. Design is another surface on the same invoice.
Figma Make, still a beta add-on in that write-up, is $20 on top, plus credits. Eigent’s table:
| Item | Cost |
|---|---|
| Image generation | 25+ credits |
| Figma Make generation | 30–100+ credits, complexity-dependent |
| Shared AI credit pool | $120–$240/month for 5,000–10,000 credits |
| Pay-as-you-go (Q2 2026) | $0.03 per credit |
March 2026 is when Figma started enforcing AI credit limits. Hit the cap and you wait, buy a pool, or go PAYG. A heavy week of Make generations is not a $20 line item. It is a credit event.
Claude’s simplicity is the pitch. Antoniou’s token complaint is the catch. Bundled does not mean infinite. If Design burns the same Pro stack you use for writing and coding, a design exploration week can starve the rest of the subscription. Max exists for that reason. It still has a ceiling he ran into.
Figma’s complexity is the pitch in reverse. You see credits. You can tell finance what a prototype costs. You can also surprise them.
If you are a solo PM on Pro, Claude Design is the cheaper experiment. If you are a 12-person product org with Figma already on the invoice, Make’s add-on is noise until the credit pool shows up. Then it is not.
Design-system fidelity, not prompt poetry
Eigent’s real battleground is the system, not the pretty first screen. Anyone can get a plausible dashboard from a paragraph in 2026. The question is whether the second screen uses the same button.
Claude Design, in the glowing version, reads your tokens and components and stays close. Antoniou agrees the import was better than a recolor. He also says it cannot replace a library with variants, constraints, and auto-layout. Style-and-pattern extraction is not a design system. It is a costume the model wears until you look at the padding.
Figma Make starts in the file that already has those variants. Point-and-edit means a designer can fix the costume without a round trip through HTML. That is why I am handing production teams the Make win. The library is already there. Generation is an on-ramp, not a second source of truth.
If you do not have a library, neither tool will invent a good one. Claude Design might look more consistent across a pitch deck because it only has to fool a slide. Figma Make on an empty file is just another generator with a Figma export already done.
Non-designers matter here. Antoniou argues Figma should worry about PMs, founders, and developers who hold seats they never wanted. Those people will try Claude Design because they already have Claude. If their wireframes never need to become a component set, Design is enough and a Figma seat is a tax. If those wireframes have to become the product, someone still opens Figma. Make at least means they open it sooner.
Junior designer panic is in that Substack too. I am not going to write the obituary. I will say this: a tool that cannot export variants is not a staffing plan. It is a first draft machine. First draft machines change who waits, not who owns the library.
Workflow: where the afternoon goes
You already live in Figma. Start in Make. Generate the flow. Copy to canvas if you are in a preview. Have a human fix spacing. Use Claude Code to Figma when the working UI already exists in a browser and you need frames, not a speech-to-mock.
You already live in Claude Code. You can push UI into Figma through MCP, per Antoniou. That path is not Claude Design. Do not buy Design because you liked a Code demo. Design is the visual surface with the HTML export problem.
You are a founder making a deck. Claude Design’s slides and one-pagers are the product. Eigent lists those on purpose. Figma Make will feel like asking a CAD tool to make a PDF.
You are handing work to engineering. Figma still wins the spec conversation at most companies. Claude Design producing runnable-ish HTML is closer to a prototype than a spec. Some teams want that. Some teams will reject anything that is not in the file with comments. Ask which meeting you are in.
Iteration speed is where Antoniou soured. First shot good, tenth shot expensive. Make’s credit meter is annoying and visible. Claude’s token burn is annoying and easy to deny until the week is gone. I prefer the visible meter if I have to explain a bill. I prefer Claude if I am alone and I already paid for Pro.
Reliability: both are previews. Eigent says so twice. Do not sign a client delivery against a preview. Use them for exploration. Lock production in the tool your design system already occupies.
A week of work, not a feature matrix
Monday: dump one real flow into both tools. Same prompt. A checkout, a settings page, a dashboard you already shipped. Save the outputs with timestamps.
Tuesday: try to match your actual buttons. In Claude Design, that means the import Antoniou liked. In Make, that means generating inside a file that already has the library. If Design “gets close” and Make uses the real component, you just learned the only lesson that matters.
Wednesday: iterate ten times on one screen. Count credits or tokens. Antoniou’s complaint will either show up or it will not. If Pro chokes, you know Max is the real price of Design. If Make asks for a credit pool, you know the $20 add-on was a teaser.
Thursday: export. Can you put Design’s HTML into the Figma file without a science project? Can you copy a Make preview to the canvas, as Figma now advertises? If the answer is a third-party converter, you do not have a round-trip.
Friday: show a designer and an engineer the same output, separately. Designers will talk variants. Engineers will ask whether the HTML is a prototype or a threat. Their answers decide the winner for your company. Mine, in the field above, is still Make for library teams.
PowerPoint is the sleeper. Antoniou likes Claude Design’s decks more than he likes fighting slides inside Figma. If your actual job this month is a board deck, stop pretending you needed Make. Use Design, export, move on. If your actual job is a design system, stop pretending a deck tool is a system.
The winner, with the caveat the field cannot hold
Figma Make for teams with a Figma library and a designer who will touch the output. That is the default professional setup, so that is the default win.
Claude Design for people whose job is not “maintain the system” and whose Claude bill is already a sunk cost. First drafts, decks, internal pitches. Export limits mean you will still paste screenshots into the real file.
If Anthropic ships a native Figma round-trip that is not “HTML plus hope,” revisit this. If Figma bundles Make into Pro and stops the credit surprise, the price argument gets quieter and the gravity argument stays.
Until then, do not switch your source of truth because a generated checkout flow looked done. It looked done. The variants are still in Figma, or they are not anywhere.
One more practical split. If your company already bought Claude Max for coding, Design is sitting there unused. Try it on the deck. If your company already bought Figma Organization, Make is sitting there unused. Try it on one flow in a branch file, not in the production library. Neither experiment requires a migration project. Both require a person who will delete the output if it is wrong. That person is the actual product. Keep them in the loop.