Summary
AI coding tools can generate a landing page within minutes, but that reveals very little about how they perform on a real-world project. I wanted a more demanding challenge, so I asked Claude Code, ChatGPT Codex, and GitHub Copilot to build the same complex finance dashboard for my upcoming home renovation.
Each tool received the exact same requirements, allowing me to compare how they planned the project, structured the code, handled UI details, and refined the overall user experience. I had a clear expectation before starting the test, but the tool that came out on top was not the one I originally predicted.
A Note on the Prompt
To ensure a fair comparison, I provided all three AI coding tools with the exact same prompt. The objective was to build a fully functional financial dashboard capable of managing my 10-month home renovation project.
The requirements included room-based budgets, contractor profiles, purchase records, pending payments, invoices, expense categories, filters, charts, and detailed reports.
I also emphasized that I wanted a working application rather than another visually polished prototype filled with placeholder buttons.
For this comparison, Claude Code used the latest Fable 5 model, ChatGPT Codex ran on GPT-5.6 Sol, and GitHub Copilot was tested inside VS Code using the latest Agents view with mai-code-1-flash, a model specifically designed for coding.
GitHub Copilot
Fast and Impressive, but Not the Winner
I began with mai-code-1-flash inside GitHub Copilot, and it was by far the fastest tool in the comparison. It transformed the prompt into a working dashboard almost instantly, making a strong first impression.
The dark blue theme looked polished, while the rounded cards gave the interface a modern SaaS-style appearance. Most of the requested sections were already available, allowing me to log expenses, review charts, monitor budgets, and get a clear overview of the renovation project with minimal effort.
However, the experience became less convincing once I explored beyond the homepage. Although the Rooms and Contractors sections displayed useful information, there was no option to create new entries or edit existing ones. Several areas felt more like static demonstrations than complete workflows.
Overall, it was a solid implementation and more than capable of producing a convincing demo, but it lacked the depth and functionality needed to compete with the stronger coding tools in this comparison.
Claude Code
Powerful Features, but an Outdated Design
Claude Code running Fable 5 was the slowest tool in the test. It took considerably longer to plan and generate the dashboard, consuming almost my entire token allowance in the process. By the end, I was approaching the five-hour usage limit, making the experience feel significantly heavier than expected.
Even so, the final result was much more complete than GitHub Copilot’s implementation. Claude Code paid close attention to the prompt and transformed most sections into fully functional workflows.
Inside the Rooms and Contractors sections, I could create new entries, edit existing records, and complete detailed forms instead of simply viewing static cards. The Expenses section was equally well organized, offering practical filters and controls that made the dashboard genuinely useful for day-to-day financial tracking.
Despite its functional strengths, the interface ultimately prevented it from taking first place. The typography felt outdated, spacing was inconsistent throughout the application, and the overall visual design lacked the polished, modern appearance I expected.
From a functionality standpoint, Claude Code was highly impressive and considerably more capable than GitHub Copilot. However, its dated presentation kept it from winning the comparison, which was surprising since I initially expected Fable 5 to lead the field.
ChatGPT Codex
Polished, Complete, Thoughtful, and Resource-Intensive
ChatGPT Codex running GPT-5.6 Sol was the clear winner of the comparison. It delivered the best balance of visual polish, feature completeness, and attention to detail.
From the moment the dashboard loaded, it looked like a finished SaaS product rather than an AI-generated prototype. The clean interface, rounded corners, consistent spacing, and well-organized layout created a professional experience throughout.
One particularly thoughtful addition was a renovation progress indicator positioned in the lower-left corner, providing a quick overview of the project’s completion status.
More importantly, Codex went beyond attractive charts and summary cards. Whether I was adding an expense, purchase, room, or contractor, every workflow included the appropriate fields needed to create detailed, information-rich records.
The final product felt practical enough to support daily use throughout an entire 10-month renovation project.
My only notable criticism involved the Codex application itself. While it was running in the background, my Mac generated noticeably more heat compared to using VS Code with Claude.
Fortunately, token consumption remained significantly lower than with Fable 5, making the overall experience more efficient despite the increased thermal load.
Even with that limitation, ChatGPT Codex delivered the most polished, complete, and practical dashboard in the comparison.
Final Verdict
This comparison reinforced an important lesson: speed, functionality, and polish rarely come together in a single package.
GitHub Copilot was exceptionally fast and produced an impressive demonstration, but its lack of complete editing workflows made the application feel unfinished.
Claude Code with Fable 5 offered the deepest functionality, yet its outdated interface, inconsistent spacing, slower execution, and high token consumption prevented it from taking the top spot.
ChatGPT Codex with GPT-5.6 Sol successfully combined a polished user interface, comprehensive workflows, thoughtful enhancements, and lower token usage than Claude, making it the strongest overall performer in this comparison.