on Codex CLI
14 messages • iMessage, Discord
thoughts re: the pros/cons of openai/codex vs. related tooling, with particular respect to [lack of] message branching
- Josh
ok I'm enjoying codex cli
fast and effective
both the model and the CLI
vs Opus in Claude Code, where both are slower and more effective in their own ways
that's to say:
- opus (in claude code) is more methodical (than gpt-5 in codex)
- CC has branching, albeit with pretty bad UX (winner by default, since no one else has any chat-branching whatsoever)
I would actually say opus feels more methodical (in part because it's so slow and the thinking that you see, I believe, is not summarized). not sure if it actually is / haven't compared enough
and again harder to compare because no | bad branching
⸺
- Josh
update (codex cli keeps getting updates, eg gpt-5 @ high reasoning effort is now easy to use in it)
⸺
- Josh
[...] no | bad branching
x.com yup
you can be "serious" about it tho, you just need to use it differently
in the same way that gemini web's absurd branching handicap necessitates that it be used differently than chatgpt
[...] no one else has any chat-branching whatsoever
and on that note + funnily enough, I realized today that the gemini cli has save states / manual checkpointing
which is in some ways better than claude code, because you can go back and forth
whereas claude code only lets you rewind; you effectively clip off further states once you jump back and lock in your fork with a msg
