Can Claude build an app? Yes, and you can try the ones it built. In Claude Code, Claude Opus 5.5 and Claude Fable 5.1 each received two app briefs, a phone finance app and a team kanban board, and built them in one attempt with nobody editing the code. Open them below and send some money or drag a card.
What did Claude build?
Two working apps per model, each from a brief that fixed the data and the features.
- The finance app. Both Claude apps open on a balance card, show the latest transactions, and run the four-step send flow: pick a contact, type an amount on an on-screen keypad, review, confirm. After sending $42.50, both update the balance to $4,783.69 and put the transfer at the top of the list. Opus 5.5 added a sparkline of the month, haptic feedback on supported phones and a full-screen confirmation; Fable 5.1 animated the balance and added money-in and money-out figures.
- The kanban board. Both Claude boards drag cards between columns with a placeholder, flag overdue cards in red, undo the last move, filter by tag, open a details panel with editable fields and a delete option, save automatically and switch to a designed dark mode. Opus 5.5's also tracks "3 of 14 done" in its header and tilts the card you are dragging. The two boards look strikingly alike, a rare case of the Claude models converging. Fable took 45 minutes, Opus 101.
How does Claude build an app?
By building, opening and fixing, in a loop. Claude Code gives the model a folder and a shell, and the Claude models use it heavily: on our briefs they ran dozens of shell commands per build, many of them loading the app in a headless browser to check what it rendered. That loop is why Claude's apps tend to arrive polished, and why they take 35 minutes or more. Claude Code is included with Claude's Pro plan, from $17 a month billed annually per Claude's pricing page.
What can't Claude build from one prompt?
The parts that are not in the browser. These apps have fixed data, no sign-in, no server and no real payments, because the briefs asked for single-file front ends. Claude Code can write backends too, but that is a longer, multi-session job with decisions you will want to make yourself. Treat these as the best kind of prototype: real, clickable, and honest about what is missing.
How does Claude compare with ChatGPT for apps?
More polish, more time, more cost. The GPT-6 models built the same finance app in 9 to 14 minutes for under $3 at list prices; Claude took 35 to 69 minutes and $11 to $21. See all four side by side in best AI for app development.
Four boards, two by Claude. Drag a card in each and see if you can tell.
Take the blind test
Frequently asked questions
Can Claude build an entire app?
A complete front-end app, yes. From one prompt each, Claude Opus 5.5 and Claude Fable 5.1 built a phone finance app with a working send-money flow and a kanban board with drag and drop, undo and filters, with no human edits. A production app with accounts and a server needs more than one prompt.
How long does Claude take to build an app?
On our briefs, 35 to 69 minutes for the finance app. Claude Code spends much of that time opening the app in a browser and fixing what it sees, which is why its first builds are polished.
Do I need Claude Code to build an app with Claude?
For a multi-step, tested build like these, yes. Claude Code comes with the Pro plan, from $17 a month billed annually. The free Claude app can write a single-file app and preview it as an Artifact, without the testing loop.
Which Claude model is best for building apps?
Claude Opus 5.5 produced the most detailed finance app, including haptic feedback, and is Anthropic's recommended default. Claude Fable 5.1 built a calmer app in half the time.