Codex mini stuck thinking

I believe it isn't stuck, it just thinks for an extremely long time. It almost always _does_ start acting again in my testing.

I believe it isn't stuck, it just thinks for an extremely long time. It almost always _does_ start acting again in my testing.

It actually gets stuck, intermittently. I have to restart codex and retry a couple times for it to work, and you know it works because it's normally very fast, < 10 seconds response turnaround including thoughts.

I also have this issue. o4-mini works but codex hangs on thinking and eventually I get a timeout message. I don’t know if it is a bug or a server side issue. It’s hard to tell. It generally happens right before I expecT a tool call to happen.

Likewise, am consistently getting 10-minute timeouts with the codex-mini model. Probably server-side, but the OpenAI status page is not showing a problem. Running with DEBUG=true does not show anything interesting:

OpenAI:DEBUG:response 200 https://api.openai.com/v1/responses e [Headers] {
  [Symbol(map)]: [Object: null prototype] {
    date: [ 'Sun, 18 May 2025 01:08:35 GMT' ],
    'content-type': [ 'text/event-stream; charset=utf-8' ],
    'transfer-encoding': [ 'chunked' ],
    connection: [ 'keep-alive' ],
...

After 600 seconds (10 minutes), this output:

    system
    ⚠️  Network error while contacting OpenAI. Please check your connection and try again.

Consistently reproducible for me.

The o4-mini model responds in a few seconds and begins making tool calls.

I really want to try out codex-mini but pretty much every time I try it gets stuck in thinking, only had it work once

I've experienced similar problems (running v0.1.2505172129). It's intermittent but sometimes it appears to reason for a very long time before ending with:

    system
    ⚠️  Network error while contacting OpenAI. Please check your connection and try again.

I can see the logs in platform.openai.com and the task actually completed successfully, including making the edits locally.

Hi everyone! I’m having the same issue with codex-mini running version 0.1.2505172129. It’s interesting that it responds to the “hello” message, but when I ask it about a file in the project, it gets stuck.

For me it's stuck too. It was thinking about 200 seconds and then I quit.

I am having the same issue, if it doesn't start executing commands within a few seconds, i have to esc esc out and try again. After one or two retries it works.

I also asked this in related issue #1032 - Can people who are reporting list the exact steps to reproduce? Including what you're using on the command line? What kind of subscriber are you? I mention this, because as a pro subscriber I get a message thanking me for being one as soon as I authenticate. I'm assuming it is tied to your subscription status/API Tier somehow.

The same thing happened to me earlier, and as soon as I authenticated the appropriate way it's been working fine for me all day long.

!Image

The current state of the codex mini model is peak bad.

Ignores instructions completely sometimes.
Thinks 5 minutes about a search and replace type of task even if you give it specific file to focus on.
Say thank you and it will think and read files for minutes.

I also asked this in related issue #1032 - Can people who are reporting list the exact steps to reproduce? Including what you're using on the command line? What kind of subscriber are you? I mention this, because as a pro subscriber I get a message thanking me for being one as soon as I authenticate. I'm assuming it is tied to your subscription status/API Tier somehow. The same thing happened to me earlier, and as soon as I authenticated the appropriate way it's been working fine for me all day long. !Image

I am a free subscriber, but using a personal OpenAI API key. Given all users are paying for API credits, I'd expect it to work the same for free vs pro subscribers?

I’m on the Tier 3 API, EU location, maybe this is a regional issue. I can confirm the mentioned issues with stuck thinking and network errors. Switching to o4-mini makes it usable again, so it’s not my local setup or network.

Same thing, Tier 4 API

in my situation it has nothing to do with login or whatever, i'm able to login with chatgpt, and i'm even able to ask codex a few questions, even had it analyze one of my repo, but when he get stuck everytime i ask him a question about a big pdf i made it eat

I have the same issue. Only codex-mini-latest has the issue and it is intermittent, roughly 40% of time, making it unusable. Switching to o4-mini and issue is gone. Most likely related to tool use. Normal conversations are fine and code reviews are also fine. I think it is related to the model, not the cli tool.

We are using it directly with the API keys, no login required. High tier with no rate limit issues.

Hope OpenAI fix it.

I think we have have established that this is (or was) an issue with codex-mini-latest making tool calls (in this thread and the duplicate thread). If enough people ping OpenAI via support, they might realize that there is a problem and do something about it.

However, it seems to be fixed for me now. In my latest conversation attempt with codex-mini, it has successfully made about 20 tool calls in a row.

I've just encountered this at least 3 times this morning. So it's not fixed.

Still not fixed. I always pull from the repository and build. So effectively I'm using the latest commit. Very inconvenient to have to esc esc all the time and repeat request multiple times before getting a response with codex-mini-latest

I'm seeing this issue too. It works for a while, then I get 'thinking...' for 10 minutes then the timeout message. I've tried 'Sign in with ChatGTP', pasting API key, and setting API key in shell, none of them seem to fix it. I'm using codex cli version 0.1.2505172129. Node 22.7.0. Debian linux host.

Same thing for me. ESCing out and retrying a couple times usually works.

This tool is kind of a joke. I could not use it even once.
First this https://github.com/openai/codex/issues/987
Now this one.

I tried with phrase "Just say hello". It gets stuck on thinking... and then, rate limit reached
Tried with ESCing and retrying, many times. Same thing.

I think that from the announcement and the recent commit history, they are abandoning the node version and instead focusing on the rust version. I've tried the rust version yesterday and the stuck issue is gone. Apparently it has to do with some streaming break and the retry handling in the node version. So for anyone looking for the solution, I'd recommend you to try the rust version. npm i -g @openai/codex@native

Thanks for the feedback! Please try again on the latest version of Codex CLI. If the issue persists, let us know.