stream disconnected before completion: Transport error: network error: error decoding response body

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

• #8294
• #8180
• #7202
• #7197
• #7198

Powered by Codex Action

We're not aware of any service outages currently. Please use the /feedback command to upload your logs and session details, then post the thread ID here.

Are you using a VPN or any network proxies?

met same problem today：Uploaded thread: 019b3921-f0df-7fa1-a471-866e632e6461 gpt-5.2-codex-high

Same issues for 5.2-codex in v0.75.0. MacOS 15.6.1
5.1-codex-mini works just fine, 5.2 high too.
019b3782-6b2b-7983-9153-18f4f91bddbb

same as 019b3842-bf50-7970-80e9-e979eb43291a

When running multiple Codex CLI processes at the same time, only one process is able to execute, and it performs very slowly. The others all fail with errors：
stream disconnected before completion:Transport error: network error: error decoding response body
However, as a Max plan user, I inevitably need to work on multiple projects concurrently.

<img width="1172" height="214" alt="Image" src="https://github.com/user-attachments/assets/39dc552e-8612-4bfe-9763-6f4dec8fef43" />

Starting from yesterday

same issue

<img width="808" height="794" alt="Image" src="https://github.com/user-attachments/assets/362fa305-eb31-4843-a099-16667511da23" />

Same here with Codex 0.77.0 and gpt-5.2-codex.

<img width="888" height="51" alt="Image" src="https://github.com/user-attachments/assets/b96d7c7a-c4ec-4ac1-823a-8882ddfd8581" />

Same today. v0.76.0 on MacOs, gpt-5.2-codex high. Completely unusable.

Same today. v0.76.0 on Ubuntu, gpt-5.2-codex high. Completely unusable.

same issue 0.76.0 MacOS. happens frequently.

Thanks all for reporting. As @etraut-openai asked, are you seeing this while using a VPN or proxy? Also, please use the /feedback command in a session where you're experiencing this issue, then post the thread ID here. Thanks.

met same problem today：Uploaded thread: 019b3921-f0df-7fa1-a471-866e632e6461 gpt-5.2-codex-high

After I switched back to gpt-5.2 (no codex suffix), it works normally in recent 2 days.

@SunRunAway Understood. Has anyone seen this issue with models other than gpt-5.2-codex? And are you using a proxy of some sort?

Same issues. vscode extension 0.4.55.Both gpt-5.2 and gpt-5.2-codex

@qinlu000 Thanks. Can you use the /feedback command in a session where the issue happened and share the thread-id? And are you using a proxy/VPN?

I have the same problem. The request stops due to this error after a few minutes of processing. Thread ID 019b4524-00aa-72e1-af67-cfbb216c9a5e.

same

Same, it seems to have happened around the time gpt-5.2-codex was released

Same 019b4a6b-a231-7353-8a2b-3285530a8663. I concur. It's happening since 5.2.

same bug, thread ID 019b4a59-6552-77a0-9a2f-c013b1b607f0

I agree, and after switching to version 5.1, it almost never happened again, so now I almost exclusively use version 5.1.

Yes, all of the reports we've received so far appear to be related to 5.2. We have a few working theories and are actively investigating.

Thanks to everyone who uploaded your logs using the /feedback command. This information has been really useful in our efforts to diagnose the problem.

I'll continue to post updates here as we make progress.

I've solved it partially by asking it to work in short chunks but it does not always obey. When it does though I get no more disconnection errors. It's only when it spins its wheels for too long that it gets disconnected. might be a GPU limit it's hitting, and the model could be doing more than it should or getting stuck in a long running thought process. My hunch.

same

I've solved it partially by asking it to work in short chunks but it does not always obey. When it does though I get no more disconnection errors. It's only when it spins its wheels for too long that it gets disconnected. might be a GPU limit it's hitting, and the model could be doing more than it should or getting stuck in a long running thought process. My hunch.

Exactly. when I let it think questions in small step, this problem disappears.

Same problem. Switching to model 5.1 fixes it.

vpn disabled and both 5.1 and 5.2 not working for me

<img width="937" height="707" alt="Image" src="https://github.com/user-attachments/assets/4f2057f3-4127-4f26-91be-d9988588a8f1" />

@LordMoMA, please use the /feedback slash command to upload your logs and session details, then post the thread ID here.

019b59fd-a88c-70c0-95cc-966a66032782
@etraut-openai

The observed behavior is limited to the 5.2 model. Switching back to version 5.1 resolves the issue and ensures smooth operation.
<img width="465" height="124" alt="Image" src="https://github.com/user-attachments/assets/2e513daa-f31b-429d-85d4-f4b7053b66aa" />

I have the same issue

<img width="1210" height="720" alt="Image" src="https://github.com/user-attachments/assets/df663d3f-8824-4388-ace9-ca58f44105ba" />

I have the same issue <img alt="Image" width="1210" height="720" src="https://private-user-images.githubusercontent.com/6851386/530343980-df663d3f-8824-4388-ace9-ca58f44105ba.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3NjY3ODMzNzYsIm5iZiI6MTc2Njc4MzA3NiwicGF0aCI6Ii82ODUxMzg2LzUzMDM0Mzk4MC1kZjY2M2QzZi04ODI0LTQzODgtYWNlOS1jYTU4ZjQ0MTA1YmEucG5nP1gtQW16LUFsZ29yaXRobT1BV1M0LUhNQUMtU0hBMjU2JlgtQW16LUNyZWRlbnRpYWw9QUtJQVZDT0RZTFNBNTNQUUs0WkElMkYyMDI1MTIyNiUyRnVzLWVhc3QtMSUyRnMzJTJGYXdzNF9yZXF1ZXN0JlgtQW16LURhdGU9MjAyNTEyMjZUMjEwNDM2WiZYLUFtei1FeHBpcmVzPTMwMCZYLUFtei1TaWduYXR1cmU9NDAzNjAyY2FlMjRlYjBmN2Q4NmI4MzRlYTkzZTY5ZTYwNTlmZTEzZGE1YjIyN2U3M2Q3NTkwM2JjOTFlZmUwNSZYLUFtei1TaWduZWRIZWFkZXJzPWhvc3QifQ.39MNWQKoXTEGSbZfsGXYiU97g6P1mmnt2ASfFfGbyjw">

Same problem ((

It's gotten much better in the last few hours.

I do have the same issue. And it seems like be getting worse.

Not sure if related, but it started yesterday when I opened another console thread with 5.2 codex model. When it was running in a single window - I didn't face any issues. But once I stopped and close the second window - the issues in the first windows remains

<img width="1160" height="70" alt="Image" src="https://github.com/user-attachments/assets/ed225e80-7697-48af-8182-cbb19b1e8231" />

I have the same problem.
Only when I use the gpt-5.2-codex model, I keep getting this error. Any other model does not cause errors.
Please, do you have any recommendations? :((( Sometimes the request works out...

macOs tahoe 26.1
thread ID: 019b79a4-0bd8-7501-8425-8a75b07c0529

<img alt="Image" width="1160" height="70" src="https://private-user-images.githubusercontent.com/58884946/531336975-ed225e80-7697-48af-8182-cbb19b1e8231.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.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.UrP9G-g2OQuKkz5vQRa3daW6IX1ioCEUAvXDBxJXlfU"> I have the same problem. Only when I use the gpt-5.2-codex model, I keep getting this error. Any other model does not cause errors. Please, do you have any recommendations? :((( Sometimes the request works out... macOs tahoe 26.1 thread ID: 019b79a4-0bd8-7501-8425-8a75b07c0529

Both gpt-5.2 and gpt-5.2-codex may cause this error, mostly occured after codex cli or vscode extension execute apply_patch to create/edit/delete files. Add "DO NOT use functions.apply_patch" to AGENTS.md of the workspace and use @modelcontextprotocol/server-filesystem@latest mcp. This may help.

please just add retries for that

Facing similar reconnecting issues with 5.1 and 5.2. Codex has been essentially unusable for me for 1-2 days with this error.

Thread ID: 019b9f87-7391-7ae2-93a0-f25ca7fa218a

````
• Reconnecting... 4/5 (16m 38s • esc to interrupt)
└ Unexpected status 502 Bad Gateway: <html>
<head><title>502 Bad Gateway</title></head>
<body>…

````

````
■ unexpected status 502 Bad Gateway: <html>
<head><title>502 Bad Gateway</title></head>
<body>
<center><h1>502 Bad Gateway</h1></center>
<hr><center>cloudflare</center>
</body>
</html>
, request id:

````

@AkanshDivker, the error you're seeing is different from the others in this thread. I don't think they're related. Could you please open a new bug report so we can track it separately? Thanks for uploading your logs.

Edit: I'm not able to retrieve your logs. I'm guessing that it failed to upload, perhaps because of the same gateway issue you're seeing.

Seeing an uptick in this, reasons for over an hour and drops the whole thing, super painful for one-shot, same for exec or in-line 019b9f3f-4139-7cf3-8fd5-05ee17b0b2e4

A temporary workaround: Switch from GPT-5.2-Codex to GPT-5.1-Codex-Max.

Test environment:
codex-cli: 0.79.0
codex-vscode-extension(Codex – OpenAI’s coding agent): 0.4.58

Hopefully the official team can identify the root cause and provide a proper fix soon.

same. thread ID 019bb10f-aab7-77a2-a7cc-5858bcc636ee. after about 20 mins.
i heard switch to 5.2 (not codex) works. also i'm on clash vpn not sure if it matters.

Hope this helps in some form, provided by 5.2 high after an autonomous session with 4 concurrent agents, 16 hours and aprox. 200m tokens:

———

We’re seeing frequent streaming disconnect/reconnect loops during long-running CLI “review” runs that heavily use tool calls (shell commands) and then continue generation.

What we observe (high-level)

• The client repeatedly logs reconnects of the form: “stream disconnected before completion: response.failed event received”, retrying up to the configured limit.

When it happens

• Reconnect sequences start overwhelmingly right after a tool call returns output (i.e., after the model receives tool output and attempts to continue the stream).
• The most common trigger pattern is large tool outputs (e.g., printing sizeable code ranges / long command output).

Where it concentrates

• The issue concentrates in the largest / highest-token scopes and in lanes that issue many tool calls (large context + many tool-output injections).
• In our run, reconnect frequency correlated strongly with total input token volume and total tool-output volume.

Why this matters

• This behavior effectively reduces safe parallelization: increasing concurrent workers increases the number of “tool-output → continue streaming” continuations in flight, raising the probability that at least one worker hits retry caps (degraded/incomplete run).

———

Same error. Thread: 019bb728-8086-76f0-befa-3f8d6181e11b

We think we have mitigated this issue. There are times when the "beefier models" (to quote @fmhall) take longer to generate the next batch of tokens during streaming, and this can sometimes exceed an HTTP timeout value.

I'm going to mark this bug as closed, but if you see this again in the next few days, please use /feedback to upload your logs and post your thread ID here.

The problem still exists.

ID: 019bc51e-a262-77d2-914e-64ea64054e0f

Same problem.

@Kritoooo, please use /feedback to upload your logs and post the thread ID here.

019bcedb-5103-78c1-95ec-b5709efa3f72

019bdf3b-39d9-7a50-bcf3-db715a50a8cf

@Kritoooo, your logs don't show a "Transport error". They show that the agent attempted to access the network but was blocked due to sandbox policy.

@Trojanking123, thanks for attempting to upload your logs. Unfortunately, they didn't come through. We had a temporary server issue that prevented some logs from uploading. We've resolved that issue now. If you're still able to repro this problem, please use the /feedback command again to upload your logs and create a new bug report. Sorry for the inconvenience.

I'm going to close this bug report thread because we think we've mitigated the original bug that was reported. If you are seeing similar behaviors, it's likely a different issue, and we'd appreciate it if you'd file a new bug report with details.

This issue has sadly returned. I've had a rough 48 hours over this, I get a ~12-15 output token delay from first input and multiple stream transport errors. I've got a ton of session ID's with this error, most recent is 019bf7ad-9cdc-7e30-a9e1-517068e0a6c4, using v0.91

@MisterRound, the underlying cause of this bug report was mitigated. If you're seeing a similar problem, then it's a new issue. Please use /feedback to upload your logs and session details and open a new bug report.