Duplicate outputs from CLI

Resolved 💬 8 comments Opened Aug 29, 2025 by haonans Closed Oct 29, 2025
💡 Likely answer: A maintainer (etraut-openai, contributor) responded on this thread — see the highlighted reply below.

What version of Codex is running?

codex-cli 0.27.0

Which model were you using?

Default, i believe gpt-5? I am a pro subscriber.

What platform is your computer?

Darwin 24.6.0 arm64 arm

What steps can reproduce the bug?

<img width="1464" height="228" alt="Image" src="https://github.com/user-attachments/assets/99f4efd3-58bf-4abe-9b50-ced6f6ec7274" />

I've been using the new codex heavily since released, it's amazing.

This issue happens not very often but I've seen a few times. IIRC, this happens usually when the line is a list item, and mostly the 1st list item.

Will add more details if I can consistently reproduce it.

What is the expected behavior?

_No response_

What do you see instead?

_No response_

Additional information

_No response_

View original on GitHub ↗

8 Comments

haonans · 10 months ago

<img width="1470" height="279" alt="Image" src="https://github.com/user-attachments/assets/7abe2f8a-01ce-47f6-ac69-ff2cf419bd7d" />

Again.

piotrmaciejbednarski · 10 months ago

I think this is a problem with the LLM model itself, not the Codex

synestiqx · 10 months ago

@piotrmaciejbednarski
It’s definitely Codex’s fault—the LLM model has nothing to do with it.

piotrmaciejbednarski · 10 months ago
@piotrmaciejbednarski It’s definitely Codex’s fault—the LLM model has nothing to do with it.

https://news.ycombinator.com/item?id=37543642

False.

synestiqx · 10 months ago
> @piotrmaciejbednarski > It’s definitely Codex’s fault—the LLM model has nothing to do with it. https://news.ycombinator.com/item?id=37543642 False.

A very valuable source of information. Very often, but it has never happened to me :) Inventing some kind of magic because someone can’t handle context—please provide me a link to a scientific article proving that such a problem exists in LLMs, because although it supposedly happens ‘very often’, it has never happened to me in any API.

piotrmaciejbednarski · 10 months ago
> > @piotrmaciejbednarski > > It’s definitely Codex’s fault—the LLM model has nothing to do with it. > > > https://news.ycombinator.com/item?id=37543642 > False. A very valuable source of information. Very often, but it has never happened to me :) Inventing some kind of magic because someone can’t handle context—please provide me a link to a scientific article proving that such a problem exists in LLMs, because although it supposedly happens ‘very often’, it has never happened to me in any API.

Were you born yesterday? Do you know what a repetition penalty is?

synestiqx · 10 months ago

@piotrmaciejbednarski
Yes, I know what repetition penalty is, and that makes repetition even less likely. Current models don’t repeat even at temperature 0. It’s an extremely rare, almost archaic case—I know this because I don’t think only in terms of yesterday, today, or tomorrow in this context. With Codex, however, it happened strikingly often, and if you had followed it, you’d know this problem is not new.
Please don’t take this as an attack — it’s just that there’s a problem in Codex with repeating output, and it would be great if this could be resolved. Honestly, in today’s world, repetition is almost an extreme.
Best regards

etraut-openai contributor · 8 months ago

I think this issue is probably no longer relevant, so I'm going to close it. If you are still seeing this issue, let us know and I'll reopen.