OpenAI's codex-cli with GPT-5 is better than Claude Code with Sonnet 4 (without thinking). Much improved
@Sauers_ its not lol
@SIGKITTEN Have you used Codex recently
@Sauers_ Has anyone tried cursor-agents with the models? How does that fare?
@noah_vandal I haven't tried because everyone says it's noticeably worse
@mark_hadid probably not idk though
@Sauers_ Gpt 5 with or without thinking?
@Mrcfyz Idk how it works but it seems to think in codex
@Sauers_ Does it have GPT or just Codex?
@sanshugoel Idk what this means
@Sauers_ Is there a YouTube video showing this
@peacenjunity Probably
@Sauers_ Since Claude Code is so good (and keeps getting better) we all forgot Codex even exists. I will give it another try but you better not be tricking us Sauers_
@Sauers_ What the fuck is "Claude Code with Sonnet 4 (without thinking)"? You can't choose between thinking/not thinking within Claude Code.
@Sauers_ This is the first time AI coding tools feel usable out of the box without tons of prompt engineering.
@Sauers_ Following this we should use Sonnet 4 with thinking then?
@Sauers_ Clause constantly over engineers things for me. I have stopped using completely. Ask it to build a hook, it will change the whole page ui lol
@Sauers_ not yet but it's getting close
@Sauers_ This is not entirely true. From my observation GPT-5 is better at planning but execution is not so great.
@Sauers_ How much does it cost though
@Sauers_ i agree, it's just that at least these features i don't think i could do without: - image pasting - toggling between planning and executing - ide integration - running bash in background with logs access - decent sdk openai needs to allocate a bit more resources
@Sauers_ Codex‑CLI might edge out Claude Code this week, but let's be real—it's like arguing over who makes the best instant noodles. Wake me up when something truly game‑changing drops.
@Sauers_ @OpenAICodexCli gpt5 actually better than @claude_code opus…the ux and feature set is much worse. But the result is much cleaner - esp for larger task (>10 files)
@Sauers_ Cc as a harness is so much better.
@Sauers_ Yep. Also the main benefit is definitely being able to tie it to your own ChatGPT account for pretty much unlimited usage for Pro. With Claude Code, I always tried to be conservative with how much I spent on the API. Maybe their Max problem would've offered a better experience.
@Sauers_ I’ve been working daily with Claude Code, Gemini CLI, and more recently Codex CLI since GPT-5 (I had also tested it with GPT-4.1/4o). Codex CLI is genuinely impressive — it integrates smoothly into my workflow and often feels like having a great analyst at hand. That said, it’s
@Sauers_ pretty satisfied w/ gpt5 actually, replacing sonnet-4
@Sauers_ how about cc with opus 4.1