Claude Opus 4.6 vs. GPT-5.3 Codex: How I shipped 93,000 lines of code in 5 days

2026-02-09 · Source: Lenny's Newsletter · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Advanced, extended

Summary

OpenAI's GPT-5.3 Codex and Anthropic's Claude Opus 4.6 coding models were rigorously tested by shipping 44 pull requests over five days, primarily focusing on a marketing site redesign and internal component refactoring. Codex, available as a desktop app, emphasizes Git primitives like repositories, branches, and worktrees, and features "skills" and "automations" for consistent task execution. However, GPT-5.2 Codex demonstrated overly literal interpretation and overfitting during the marketing site redesign, struggling with creative, broad tasks. Claude Opus 4.6, tested within the Cursor desktop app, excelled at generative, greenfield work, producing a superior site redesign and effective initial refactoring. A combined workflow emerged where Opus handled initial development (80-90% completion), and Codex performed architectural review, bug detection, and code polishing, replicating a principal software engineer's role.

Key takeaway

For AI Architects and VP of Engineering evaluating new coding models, consider a dual-model strategy. Deploy Claude Opus 4.6 for initial product development and feature implementation, leveraging its strength in generative and creative tasks. Subsequently, integrate GPT-5.3 Codex for rigorous code review, architectural validation, and identifying edge cases, treating it as a "principal engineer" for hardening code quality. This approach can significantly accelerate development cycles and improve code robustness, despite the higher token costs associated with advanced models.

Key insights

Combining Claude Opus 4.6 for generative coding with GPT-5.3 Codex for code review optimizes development workflows.

Principles

• Models overfit to literal prompts.
• Generative models excel at greenfield work.
• Review models identify architectural flaws.

Method

Use Opus 4.6 for initial feature implementation and broad redesigns, then employ Codex for architectural review, bug detection, and code polishing before shipping to production.

In practice

• Redesign marketing sites with Opus 4.6.
• Refactor complex components using Opus.
• Use Codex for final code review and bug fixes.

Topics

• AI Code Generation
• GPT-5.3 Codex
• Claude Opus 4.6
• Software Development Workflows
• Code Review

Best for: AI Architect, CTO, VP of Engineering/Data, AI Engineer, Machine Learning Engineer, Software Engineer

Related on AIssential

• Claude Opus 4.6 vs GPT-5.3 Codex: How I shipped 93,000 lines of code in 5 days — Pairing generative AI models with specialized review models optimizes complex software development workflows.
• GPT-5.3 Codex Is INSANE! OpenAI’s BEST Model Might Beat Opus 4.6? (Fully Tested) — GPT-5.3 Codex offers enhanced speed, autonomy, and full software lifecycle support for coding and knowledge work.
• Opus 4.6 vs GPT-5.3-Codex — AI development is rapidly advancing towards multi-agent environments and parallelism, enhancing model capabilities.
• Claude Opus 4.8: where it excels and where it falls short — Claude Opus 4.8 excels at initial code generation and architectural adherence but struggles with the final 10% of complex tasks.
• 🎙️ This week on How I AI: Opus vs. Codex showdown, and AI for accessibility — Combine generative and review-focused AI models for optimal software development workflows.
• Claude Opus 4.8: Best AI Model Ever? Powerful, Agentic, and Faster! (Fully Tested) — Claude Opus 4.8 offers incremental improvements in agentic workflows and honesty, excelling in raw output quality but lagging in efficiency.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Lenny's Newsletter.