Q/A & AI Agent Showdown: Claude Sonnet 4.5 vs GPT-5-Codex

Morning build session: I put Claude Sonnet 4.5 and GPT 5 Codex head to head on the same website case requested by

Claude one-shotted the full site (frontend + backend, all wired up with blog post support) in a single prompt. It was legitimately impressive. Also Sonnet 4.5 performed this twice as fast as it took Codex end to end.

Codex still got the job done (it required 3-4 follow up prompt to get everything working and polished), but both shipped solid first drafts.

So is Sonnet 4.5 an improvement from my testing, and will i be switching back to Claude from Codex?

Two things are clear:

Tool speed and “autonomy” are now material differences, Sonnet 4.5 felt significantly more confident with tool use compared with previous models and Codex. And it was blazingly fast (5m to complete vs the 15m Codex took)

The gap is closing. 6 Weeks ago Claude Opus 4.1 was the best model, Then GPT 5 Codex dropped and out performed it, and now Sonnet 4.5 is out performing. The “best model” is shifting quickly.

I moved from GPT-5 Codex to Codex just 2 weeks ago, and now Sonnet is making me question it. The AI stack is more volatile than ever.

If you’re navigating the current AI stack or new to building with Codex or Claude, drop your questions in the comments below and I’ll help however I can.

Stay sharp,

Vince

P.S: Links to the sites and codebases are in the top comment!

3 comments

Q/A & AI Agent Showdown: Claude Sonnet 4.5 vs GPT-5-Codex