[CLI/macOS] Frequent reconnect loops and "stream disconnected before completion" errors even for simple prompts
What version of Codex CLI is running?
codex-cli 0.147.0
What subscription do you have?
ChatGPT-Pro
Which model were you using?
gpt-5.6-sol, gpt-5.6-luna
What platform is your computer?
_No response_
What terminal emulator and version are you using (if applicable)?
zsh: command not found: Terminal.app
Codex doctor report
What issue are you seeing?
I am using Codex CLI on macOS, and since a few hours ago on August 9, 2026, I have been experiencing frequent and persistent connection/streaming failures.
The most important characteristic of this issue is that it is intermittent. Codex does not fail completely all the time — it repeatedly switches between working normally and failing.
For example, Codex may enter a Reconnecting... loop for some time and then suddenly recover and successfully generate a response. However, after working normally for a short period, the next prompt may trigger the same reconnect loop again and eventually fail with a stream disconnection error.
In other words, the behavior repeatedly looks like this:
Working normally
→ Reconnecting...
→ Reconnecting...
→ suddenly recovers and responds successfully
→ works normally for a short time
→ Reconnecting...
→ eventually fails with "stream disconnected before completion"This can happen even with extremely simple prompts, such as sending a one-line greeting like "hello." It does not appear to be related to prompt complexity or long-running tasks.
The following error occurs frequently:
stream disconnected before completion: An error occurred while processing
your request. You can retry your request, or contact us through our help
center at help.openai.com if the error persists. Please include the request
ID 565eaf1d-c507-4601-a9fe-cab9c8333c7c in your message.I also repeatedly see reconnect attempts like this:
Reconnecting... 1/5 (3m 34s • esc to interrupt)
└ Stream disconnected before completion: An error occurred while processing
your request. You can retry your request, or contact us through our help
center at help.openai.com if the error persists.Sometimes Codex successfully reconnects during these attempts and resumes responding normally. However, the connection becomes unstable again shortly afterward, and the same reconnect/error cycle repeats.
This makes the issue particularly confusing because the service appears to recover temporarily, but the recovery is not stable.
Codex CLI was working normally before this started. The problem began suddenly a few hours ago and has continued since then.
I am also seeing multiple reports from other users describing similar reconnecting and stream disconnected before completion symptoms, so this may not be isolated to my local environment.
Please investigate whether there is an intermittent issue with the Codex backend, streaming infrastructure, connection handling, or request routing.
What steps can reproduce the bug?
Uploaded thread: 019fe40f-9f03-70c3-bfb7-a930d5d51ce2
What is the expected behavior?
Codex CLI should maintain a stable streaming connection and complete responses normally without repeatedly entering reconnect loops.
Simple prompts should receive responses immediately without repeated Reconnecting... attempts or stream disconnected before completion errors.
Additional information
The issue started suddenly a few hours ago on August 9, 2026.
Codex CLI had been working normally before this.
The problem occurs repeatedly and makes Codex CLI effectively unusable because even simple conversations can trigger multiple reconnect attempts lasting several minutes.
I am seeing reports from other users describing similar symptoms as well.
Additional request ID:
565eaf1d-c507-4601-a9fe-cab9c8333c7c
7 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
I can reproduce what appears to be the same issue on macOS, with an additional model A/B control that may help isolate the problem.
Environment / reproduction matrix
Minimal reproduction
Additional isolation already performed
The original failure was first observed during a more complex agentic task, so I tested whether prompt complexity or thread size was the cause. The failure still occurs on a brand-new thread with the minimal read-only prompt above.
I then changed the Mac from the normal Wi-Fi connection to a mobile hotspot. GPT-5.6 still failed, which makes a local router/Wi-Fi-specific issue less likely.
The same GPT-5.6 failure is reproducible in both Codex Desktop and the VS Code Codex extension, while GPT-5.5 works in the same environment. This makes the issue look model/route/stream-specific rather than repository-, prompt-, thread-, IDE-, or ordinary network-specific.
Request IDs
25ac8017-bc4f-4836-b391-4e00718a512e7d4398b2-eaed-44d0-905c-7f9743cefaa7Observed on August 9, 2026 (KST).
I can provide additional sanitized diagnostics if maintainers need a specific Codex log or doctor output.
Follow-up / correction to my previous comment: the issue is not limited to GPT-5.6.
I have now reproduced the same failure while using GPT-5.5 Extra High / xhigh.
New request ID:
afc6a654-4b8a-4b36-b527-0c31db5499c3This changes the isolation result. Earlier, GPT-5.5 had worked while GPT-5.6 failed, but GPT-5.5 xhigh has now also produced the same generic:
So the stronger current hypothesis is a broader intermittent Codex Responses/streaming/backend issue rather than a GPT-5.6-only regression.
Relevant controls already performed remain:
There are also existing reports of similar failures with GPT-5.5 xhigh / Extra High (for example #19440 and #30933), so this may be part of the broader streaming/reconnect failure class rather than a single-model problem.
Observed on August 9, 2026 (KST).
Additional reproduction on August 9, 2026 (KST): the same failure occurred again while using GPT-5.5 Extra High.
New request ID:
70d03076-ee8c-45d8-9ca6-305b51ac1221This is another independent request ID after the previously reported failures. At this point the issue is repeatedly reproducible across multiple turns and is not isolated to GPT-5.6; GPT-5.5 Extra High is also affected.
Additional reproduction update:
GPT-5.5 High now also fails with the same
stream disconnected before completionerror.New request ID:
1cbfd661-6b34-40e2-be55-849c18fb551eThis further weakens the hypothesis that the issue is limited to GPT-5.6 or only to GPT-5.5 Extra High / very long reasoning turns.
Current observed matrix on the same macOS environment:
The failure remains reproducible across Codex Desktop and VS Code, fresh threads, simple read-only prompts, normal Wi-Fi, and mobile hotspot.
If GPT-5.5 Medium also fails, I will add the request ID and update the matrix again.
Additional reproduction: GPT-5.5 Medium also failed with the same
stream disconnected before completionerror.New request ID:
d3537c48-6e5e-4dca-8702-142286c6570dThis further weakens the idea that the issue is limited to GPT-5.6 or to higher reasoning levels. At this point I have reproduced the same failure on:
across Codex Desktop and VS Code, including fresh threads and simple read-only prompts, and on both normal Wi-Fi and a mobile hotspot.
Observed on August 9, 2026 (KST).
same here