GPT 5 & Claude 4 Sonnet: What is the Best AI Coding Model?

Summary

This video tests OpenAI's GPT-5 against Cloud Force Sonnet in generating SwiftUI code for two projects: a drawing app and a realistic fireworks animation.Transcript summary generation GPT-5 produces a fully functional drawing app with an elegant, movable PencilKit UI, while Cloud Force Sonnet delivers a sidebar-based UI that works but requires multiple error fixes. For fireworks, both models create functional tap-trigger animations, with Cloud Force offering more customization and GPT-5 providing smooth performance. Overall, GPT-5 excels in UI polish and fewer manual corrections, whereas Cloud Force is faster and feature-rich, demonstrating the strengths and trade-offs of both AI code generators.

Takeaways

• 🖥️ The video tests GPT-5 and Cloud Force Sonnet on generating SwiftUI code for two projects: a drawing app and a fireworks animation.
• ✏️ For the drawing app, GPT-5 generated a fully functional app with PencilKit tools including pencil, marker, eraser, ruler, undo/redo, and drag-and-drop tool placement.
• ⚡ Cloud Force Sonnet generated the drawing app faster but required multiple error fixes, and its UI was less polished compared to GPT-5.
• 📱 GPT-5’s drawing app UI was more consistent across devices, while Cloud Force Sonnet’s version displayed better on iPhone than iPad.
• 🎆 Both models successfully generated fireworks animations triggered by taps, but Cloud Force Sonnet offered more customization options.
• ⏱️ GPT-5 took slightly longer to generate the fireworks animation code but produced a stable, functional result.
• 🔧 Cloud Force Sonnet required iterative fixes to handle errors in the generated drawing app code, showing less stability than GPT-5.
• 💻 In Xcode, GPT-5’s generated code had minor warnings but ran without errors, while Cloud Force Sonnet initially caused multiple errors.
• -
• 📊 Overall, GPT-5 excels in UI polish and stable SwiftUI integration, whereas Cloud Force Sonnet prioritizes speed and customizable features.
• -
• ✅ Both models demonstrate strong capabilities in generating working SwiftUI apps, highlighting their usefulness for rapid prototyping and app development.

Q & A

What was the main purpose of the video?

The main purpose of the video was to compare GPT-5 and Cloud Force Sonnet by testing their ability to generate SwiftUI code for two tasks: a drawing app and a fireworks animation.

How were the tests structured in the video?

The tests were structured by creating two sets of files in Xcode—one for GPT-5 and one for Cloud Force Sonnet—and using the same prompts to generate code for the two tasks. Each model's output was then compared and tested in Xcode.

What tools did GPT-5 generate for the drawing app?

GPT-5 generated a drawing app using the standard PencilKit UI, which included tools like a pencil, marker, eraser, undo/redo options, a ruler, and the ability to share and hide the drawing tools.

What were the key differences in the UI design between GPT-5 and Cloud Force Sonnet for the drawing app?

GPT-5's UI was praised for its elegance, offering a more polished design with the tools easily accessible and draggable on the edges of the device. Cloud Force Sonnet placed all the tools in a sidebar, which was functional but less visually appealing.

Did GPT-5 or Cloud Force Sonnet handle errors better in the drawing app?

Cloud Force Sonnet initially had more errors in its output, but it was able to fix them after some modifications. GPT-5 generated fewer errors but had a more polished overall output.

How did the fireworks animation differ between GPT-5 and Cloud Force Sonnet?

Both models generated a fireworks animation that could be triggered by tapping the screen. Cloud Force Sonnet offered more customization options for the animation, while GPT-5's version was functional but with fewer customization features.

Which model generated the fireworks animation faster?

Cloud Force Sonnet generated the fireworks animation faster than GPT-5, which took longer to process the task.

What happened when Cloud Force Sonnet couldn't fix the errors in the drawing app code automatically?

When Cloud Force Sonnet couldn't fix the errors automatically, the user manually copied the errors and fed them back to the model to apply the fixes, ultimately updating the code and resolving the issues.

Why was GPT-5 considered better for creating the drawing app UI?

GPT-5 was considered better for creating the drawing app UI because it provided a more polished and user-friendly design, with the drawing tools presented in a more aesthetically pleasing and functional manner.

How did the two models compare in terms of functionality for both the drawing app and fireworks animation?

Both models were capable of generating functional versions of the drawing app and fireworks animation. However, GPT-5 was better for UI design and visual polish, while Cloud Force Sonnet was faster in generating code and provided more customization options for the animation.