The rise of vibe coding is based on the promise of services like GPT-5: that in the future, you won’t have to know how to program at all in order to “create” software — you’ll just need to know how to communicate your software ideas in plain old English. The AI will do all the programming and debugging for you.
The ultimate goal: “software on demand,” or any kind of app, website, or visualization you can dream up, whenever you want it. But how close are we? If you’re totally new to coding and don’t want to learn how to code yourself, should you expect to get something usable? I set out to try some of the bigger consumer AI services to see how they perform. In short, I’m looking for the best vibe coding tools for beginners, not more advanced tools like Cursor or Windsurf.
For these tests, I used GPT-5, Claude 4.1 Opus, and Grok 4 (Expert).
How I tested vibe coding
To see how well each of the different AI services performed when it came to coding, I asked each service to create a web-based game, a web-based productivity service, and an iOS app.
Just like when I tested the best AI image generators, I used the same prompts to test each model. Here are the vibe coding prompts I used:
-
Make a 2D browser game where you control a spaceship with the arrow keys and shoot lasers with the spacebar. Asteroids should spawn randomly and drift across the screen. If a laser hits an asteroid, it explodes and your score goes up. If the spaceship hits an asteroid, the game ends and shows the final score.
-
Make a web-based to-do app that lets me add things to my list, view items as a list or in a calendar, and lets me add information like a time it needs to be completed (not just a date). Also, add a random quote generator that shows a different random inspirational quote each time I press a button.
-
Create a complete iOS app for tracking personal fitness goals, where users can log daily workouts and view progress charts. The app should include a home screen showing a list of logged workouts with details like date, type, duration, and calories burned, plus options to add new entries, edit or delete existing ones, and search or filter by date or workout type.
My prompts test different kinds of software and services, as well as different kinds of information. The AI will need to build some kind of physics framework for the game, for example, and either find a pre-set list of quotes, or call an API to find quotes, in the productivity app.
I’m certainly no expert coder. I do have some basic web-development knowledge and can find my way around HTML and CSS files just fine, but that’s about where my experience in programming ends. I certainly won’t be able to assess how these services perform when it comes to code efficiency. I’m judging the end product based on how well those services follow my prompts.
I’ve included links to the game and webpages these vibe coding tools created, but I won’t ask you to download the app. You’ll just have to take my word for it.
Vibe-coding a Web-based game
Credit: Screenshot: Christian de Looper
See Play the results:
All three services were able to create a working web game, with slightly different mechanics and additional touches. ChatGPT and Grok offered a game that included asteroids flying in from all directions and the ability to rotate the ship. Claude went for a Space Invaders-type set-up, with asteroids only entering from the top of the screen, and the ship pointed at the top.
Still, Claude’s approach was the best-looking. There were bright, retro colors, a background with stars, and a cool movement trail behind the user. ChatGPT came in close second, and while there weren’t stars in the background or a sense of movement, I liked its minimalistic approach, which highlighted explosions. Grok’s game looked the worst — and played the worst — the design was boring and the asteroids were just gray circles.
Credit: Screenshot: Christian de Looper
Credit: Screenshot: Christian de Looper
In the end I preferred the overall look of Claude’s game, but the functionality of ChatGPT’s.
Winners: ChatGPT, Claude
Vibe-coding a Web-based app
Credit: Screenshot: Christian de Looper
See the results:
All three services were able to create a productivity app that allowed the user to see the tasks on a calendar, and to see inspirational quotes with the press of a button. The basic interface was similar across all three of the services, too. They all opened to a list view by default, offering fields to type in a task and an option to select a date and time for the task.
However, the apps differed greatly in appearance. Grok was the most basic, once again. Its app looked like a Word document, with a few buttons and fields to enter text. Once again, ChatGPT was a significant step up in look, but not as flashy as Claude. It gave the inspirational quote a different background color, while Claude gave the whole interface a different color and used emoji to highlight times and dates.
Credit: Screenshot: Christian de Looper
Credit: Screenshot: Christian de Looper
Differences in functionality were even greater. Grok’s app was basically unusable; once a task was created, there was no way to remove it or check it off. Claude included a “delete” button next to tasks in the list view; clicking on a task in the calendar view generated a prompt asking if you wanted to delete a task.
ChatGPT, however, was the most functional in key areas. In the list view, it included a checkbox to mark a task as completed, along with a separate button to delete a task, and even a button to edit a task. It also included a “Notes” field for each task, allowing users to add more than just the task name and its due date. All of which for a much better to-do list, even if it wasn’t the flashiest.
Winner: ChatGPT
Vibe-coding an iOS app
This one was…a little more challenging.
Why? Well, Claude is the only service that actually generated a working iOS app. Both ChatGPT and Grok created apps that failed to run in Apple’s Xcode. The app that Claude created ran fine.
That said, Claude didn’t really go the extra mile.
Credit: Screenshot: Christian de Looper
Credit: Screenshot: Christian de Looper
Claude’s app allowed me to enter workouts, then displayed them in a list view, allowing me to filter based on workout type, time they were logged, and so on. But it didn’t calculate calories burned for me (I had to enter that information just like I entered workout types and how long I worked out for). And its “progress” chart was just a bar graph that showed how many calories I burned during different workout types. There was no way to create or view goals.
To actually run an app on your iPhone that isn’t from the App Store, you have to jump through some hoops. This isn’t something beginners can do in a minute or two; they’d be better off sticking to web-based AI coding for now.
Winner: Claude
The state of vibe coding
Two things are clear. Firstly, AI services are getting very good at helping non-coders or beginner coders create apps, services, and games. Secondly, we cannot yet dream up any piece of software and expect an AI service to spit out a flawless, working product. I don’t think we’re very far off, however.
App creation takes a little more than just generating a downloadable file. if you’re not interested in troubleshooting or learning any basics, I’d steer clear from trying to create actual iOS or Android apps. Web-only apps are preferable.
So what should you use to code if you’re a beginner? Well, if all you want to do is experiment with web apps and little games, both ChatGPT and Claude work well. Excluding the app creation test, ChatGPT performed better. OpenAI’s GPT-5, being free to all, is easier to access than Claude Opus 4.1 — and for those who subscribe to each service’s respective $20-per-month plan, OpenAI imposes fewer limits. Still, don’t expect sub-par coding from Claude.
And to be fair to Grok, you’ll still likely get working websites if you use it — they’ll just be basic, boring, and may not stick as closely to your prompts. As with Grok Imagine, xAI has a long way to go to catch up with its rivals.
Disclosure: Ziff Davis, Mashable’s parent company, in April filed a lawsuit against OpenAI, alleging it infringed Ziff Davis copyrights in training and operating its AI systems.
