AI Coding Showdown: Developer Ships 93,000 Lines of Code in 5 Days with Claude Opus and GPT-5.3 Codex

The rapid evolution of AI is transforming software development, and a recent experiment provides a compelling glimpse into this future. A developer undertook a head-to-head comparison of Anthropic's Claude Opus 4.6 (and its 'Fast' variant) and OpenAI's GPT-5.3 Codex, tasking them with redesigning a marketing website and refactoring complex code components. The results, detailing the creation of 44 pull requests containing 98 commits across 1,088 files in just five days, showcase the potential of these AI tools to significantly boost developer productivity. The individual published these findings as both a blog post and companion video available on YouTube, Spotify, and Apple Podcasts.

The experiment focused on identifying each model's strengths. The developer found that GPT-5.3 Codex excelled at code review and architectural analysis, leveraging its understanding of Git primitives, skills, and automations. However, it struggled with more creative, greenfield development tasks, sometimes exhibiting a too-literal interpretation of prompts. On the other hand, Claude Opus 4.6 demonstrated greater aptitude for creative development and refactoring tasks, producing visually appealing results in the website redesign challenge.

The developer emphasized the surprising synergy between the two models. While Codex proved reliable for scrutinizing existing code and suggesting improvements, Opus 4.6 shone in generating new code and tackling complex refactoring projects. By strategically combining the two, a more effective AI-powered engineering workflow was achieved.

Cost considerations are also important. While Opus 4.6 Fast offers increased speed, it comes with a significant price increase (reportedly 6x). Developers need to carefully manage their token budgets to avoid unexpected expenses. The experiment underscored the value of using Git concepts like work trees to streamline workflows and maximize the productivity gains offered by AI coding assistants.

This real-world test case demonstrates the potential of AI to augment software engineering. By understanding the strengths and weaknesses of each model, developers can create powerful AI-assisted workflows that accelerate development cycles, improve code quality, and free up human engineers to focus on higher-level tasks. The future of coding appears to be a collaborative one, where humans and AI work together to build better software faster.