For the love of god - no MD or seeing uncollapsed thinking. PLEASE!
What feature would you like to see?
OK I feel like this should be very straightforward and I hope that you guys are thinking about that but from my experience and having used Claude code I've noticed that I think the thinking mode shows up like we know the model is thinking, but we don't have to see all of it you know in fact, I would prefer to not see thinking in a CLI environment. It's just not helpful. You could do one thing which is like maybe clicking on something like a control c where it will show you the thinking breakdown, but I would rather not see that. I just wanna see the answer. Also, I don't wanna see mark down because it doesn't look good in the CLI you guys should learn from Claude code. This is very embarrassing because this shouldn't be too hard to do honestly with all these resources that you have. I don't understand it but I'm doing you a huge favor for user experience.
Love you. Hate you.
I'm happy to help :)
Are you interested in implementing this feature?
_No response_
Additional information
_No response_
11 Comments
Try minimal reasoning - really helped me here.
how do you do that ?
You can do it like using config like this
I agree, it should be minimal by default.
Unfortunately this doesn't work with GPT-5, only two options are possible:
detailedornone.I hope this will change, I opened a bug for it #2376
I don't agree with removing markdown. It wasn't working well before, but after 0.22.0 it's rendering perfectly and makes the output way easier to read.
I agree tho with removing some weight from the "thinking" output. Ideally it could be shown/hidden like in claude code, but at the very least it could use a different color with less contrast, like grey.
I’m confused why we need markdown here. Markdown is useful where it renders not in a CLI environment. Honestly there’s no excuses. Codex works well right now we just need to tighten up the formatting a bit make sure we leave thinking alone but keep it hidden. Auto is actually great - maybe with options for those who want it. Claude code does this I don’t see why these suggestions with all due respect make any sense. OpenAI needs to step up this is not too difficult. I would help but I have my own projects right now.
Get Outlook for iOS<https://aka.ms/o0ukef>
[cid:logo_34b92a0d-421e-419d-a8e9-8570bab7034c.png] <https://www.octegenmusic.com/>
Resources
(888)-OCTEGEN<tel:(888)-OCTEGEN>
@./> CLICK HERE TO BOOK YOUR FREE CONSULTATION @./>
________________________________
From: Nicolas Laurito @.*>
Sent: Saturday, August 16, 2025 12:21:01 PM
To: openai/codex @.*>
Cc: Resources @.>; Author @.>
Subject: Re: [openai/codex] For the love of god - no MD or seeing uncollapsed thinking. PLEASE! (Issue #2375)
[https://avatars.githubusercontent.com/u/5358689?s=20&v=4]Zamaroht left a comment (openai/codex#2375)<https://github.com/openai/codex/issues/2375#issuecomment-3193859280>
I don't agree with removing markdown. It wasn't working well before, but after 0.22.0 it's rendering perfectly and makes the output way easier to read.
I agree tho with removing some weight from the "thinking" output. Ideally it could be shown/hidden like in claude code, but at the very least it could use a different color with less contrast, like grey.
—
Reply to this email directly, view it on GitHub<https://github.com/openai/codex/issues/2375#issuecomment-3193859280>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/BKEK54K55WW34UIFDZQ63X33N6AB3AVCNFSM6AAAAACEBC4D62VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTCOJTHA2TSMRYGA>.
You are receiving this because you authored the thread.Message ID: @.***>
Version
0.24.0added aCtrl+Ttranscript viewer mode, and hides thinking by default (it can be viewed in the transcript):---
There is also a model verbosity config for GPT-5 since version
0.24.0:---
@cyruszad As far as I have seen, it is rendered in the CLI environment?
As @Zamaroht pointed out in the message before yours:
I think they updated it, but it looks like they overcorrected. Now they basically don't show any thinking at all, unless you go in the transcript, and there it is intermixed with code reads which is super noisy.
I think I would be fine with this change, but I think they should remove the code reads from the transcript because I don't see how they are useful. Maybe just keep a one-line like _Read from App.tsx from line 10 to 123_, or whatevr
@sldx I think they are useful sometimes in that they provide transparency for the context the model was seeing; but I agree that in a lot of standard usecases they end up being overly verbose.
I think it would be nice if the
Ctrl+Ttranscript viewer allowed 'messages' / 'sections' of the transcript to be 'collapsed'; that way longer code view sections could be collapsed by default, but we could select them and expand it if we really wanted to see what was in them.Though I suspect creating a more focussed issue about that for visibility would be better than it being lost in these comments.
---
Edit: Created a new issue for this:
I noticed the change. BRAVO. Tbh I don't care if we see the thinking. It would be useful. But I would focus more on the existing bugs we have now first. FOR EXAMPLE: the constant having to give permissions to allow it to do stuff - again Claude Code doesn't have this issue - I actually moved from CC. This is getting there. Would be cool if we could pass secrets or set a file with our passwords for sudo commands that would respect privacy but help bring this to the next level. More context control - like memory integration besides the agents.md file. We don't need all the features - we just need the features that are useful and make sense. Over-engineering doesn't help either. Also, I noticed GPT-5 has issues with being straight to the point. Too verbose. I mean it is a hard job, but I think they over-engineered it unfortunately. Anthropic's less is more motto is KEY. Less features - more reliability and accuracy PLS. This should be paid advice but then again, I won't charge for my 5 mins. You're welcome!
[cid:logo_34b92a0d-421e-419d-a8e9-8570bab7034c.png] <https://www.octegenmusic.com/>
Resources
(888)-OCTEGEN<tel:(888)-OCTEGEN>
@./> CLICK HERE TO BOOK YOUR FREE CONSULTATION @./>
________________________________
From: cosin @.*>
Sent: Friday, August 29, 2025 2:34 PM
To: openai/codex @.*>
Cc: Resources @.>; Mention @.>
Subject: Re: [openai/codex] For the love of god - no MD or seeing uncollapsed thinking. PLEASE! (Issue #2375)
[https://avatars.githubusercontent.com/u/1038751?s=20&v=4]sldx left a comment (openai/codex#2375)<https://github.com/openai/codex/issues/2375#issuecomment-3238336044>
I think they updated it, but it looks like they overcorrected. Now they basically don't show any thinking at all, unless you go in the transcript, and there it is intermixed with code reads which is super noisy.
I think I would be fine with this change, but I think they should remove the code reads from the transcript because I don't see how they are useful. Maybe just keep a one-line like Read from App.tsx from line 10 to 123, or whatevr
—
Reply to this email directly, view it on GitHub<https://github.com/openai/codex/issues/2375#issuecomment-3238336044>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/BKEK54M25G4JFF5CH55APIT3QDBOBAVCNFSM6AAAAACEBC4D62VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTEMZYGMZTMMBUGQ>.
You are receiving this because you were mentioned.Message ID: @.***>
Streamed thinking states should be much cleaner now than it was when this issue was first created.