Claude Code vs Codex: 6 Months In, Here's My Honest Split

Most "Claude Code vs Codex" posts are benchmark drag races. Useful, but they miss the thing you actually feel when you use both daily. They're good at different jobs. So here's the honest version. My real split, and why it's landed where it has.

My split, and how it moved

For about six months it wasn't a contest. I was 100% Claude, 0% GPT for coding. Then Codex shipped a version that genuinely won me over around May. It earned a real slice of my work, first about 20%, and now closer to 40%, depending on the task.

The pattern underneath that number is simple. I reach for Claude when the work involves taste and discernment. Design, naming, refactors, anything where "is this actually good?" is the hard part. I reach for Codex when I'm shipping a well-defined thing on a surface it's strong at. It takes a tight spec and grinds it out, sometimes better than Claude on that specific layer.

That's not me being diplomatic. It maps almost exactly onto what the wider data says.

The commodity facts (June 2026)

Strip away the tribalism and the public numbers tell a consistent story:

*A widely-cited Express.js refactor comparison. One data point, not a law. But the direction (Codex cheaper, Claude cleaner) shows up everywhere.

So which should you use?

The boring, correct answer the whole industry has landed on: most heavy users run both. They're cheap relative to the leverage, and they're good at different things. If you're choosing where to start:

• Start with Claude Code if you're newer, doing design- and taste-heavy work, or you want a tool that explains itself and asks before doing something risky. This is also the better first stop for non-developers.
• Add Codex once you can write a tight spec and want something shipped fast, cheaply, and unattended.

Still deciding for a specific job? Hand the decision to the AI you already have:

I'm deciding between Claude Code and Codex for this task: [describe the task].
Given that, as of mid-2026, Claude tends to win on code quality, refactors, and judgment-heavy work, and Codex tends to win on speed, autonomy, and cost for well-scoped tasks, which would you recommend for THIS task, and why? Be specific about the trade-off.

Where I think this goes

Do I see Codex closing the gap on Claude? Yes. I'd call the trajectory closer to 50-50 than my current split suggests. But the race between these two isn't the interesting part.

The open-source models are roughly six to eight months behind the frontier, and they're catching up. I would gladly run a capable model on my own machine instead of paying big tech every month. So the shift I actually expect isn't Codex overtaking Claude. It's a third engine joining the split: a local, non-frontier model handling the repetitive daily tasks, with the frontier tools saved for the work that genuinely needs them. Same instinct as the rest of this stack. Own more, rent less.