Codex responds too slowly

[ty91]

What issue are you seeing?

I'm using Codex via the Codex CLI and also the Pi CLI. Responses have been coming too slowly since last Saturday, I think, especially when it falls back to SSE from WebSocket.

What steps can reproduce the bug?

Configure the Codex CLI via ~/.codex/config.toml to use SSE explicitly, like:

model_provider = "openai-sse"

[model_providers.openai-sse]
name = "OpenAI"
wire_api = "responses"
requires_openai_auth = true
supports_websockets = false

Then start a new session. Codex takes at least 5 minutes to answer.

What is the expected behavior?

No response

Additional information

No response

[GGBondBlueWhale]

I got the same issue in these few days.

[github-actions]

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

• Recoverable reconnect loop before HTTP fallback when entering new or resumed thread #24045
• WebSocket fallback should recover after transient transport failures #23665
• Codex CLI intermittently gets stuck in “Working” state; reconnect retries occur before normal execution #24419
• Codex IDE extension often stays on "Thinking" for a long time even for simple prompts #24414
• GPT-5.5 Fast suddenly feels as slow as Standard, with long thinking/context/search stalls #24422

Powered by Codex Action

[ty91]

@GGBondBlueWhale Yep, looks like there are a bunch of open issues already. I ran into this one while using Pi, and it seems the Codex backend's SSE connection has some problems.

[Yorick-Ryu]

same issue

[OverHash]

This appears to be linked to anomalyco/opencode#29079. OpenAI's SSE endpoint appears to not respond well -- sometimes leaving an empty response, especially for oceania users.

[adeperio]

Yep confirm that I'm seeing the issue on Opencode too

[MisterRound]

See my issue, it's the same #24260

[DimaSergeew]

I encountered the same problem

[jordan584]

I got the same problem in codex app

[jordan584]

I can resolv this problem， Maybe you use vpn, and then you have five reconnecting, after that, is thinking. but it is too slow,
so you need to use TUN Mode clash. It can work to me