codex-cli on WSL: ChatGPT login works but API key mode fails with error sending request for url (https://api.openai.com/v1/responses)

What version of Codex is running?

codex-cli 0.57.0

What subscription do you have?

Pro

Which model were you using?

any

What platform is your computer?

Ubuntu 24.04.1 LTS on WSL2

What issue are you seeing?

When using API key auth in Codex CLI on WSL, every call to the model fails with:

error sending request for url (https://api.openai.com/v1/responses)

The CLI retries a few times and then the conversation effectively times out / stalls.

However:

• The same API key works from WSL with plain curl against /v1/responses.
• Codex CLI on Windows PowerShell works with an API key.
• Codex CLI on WSL works if I sign in with ChatGPT instead of using an API key.

So the bug is specific to Codex CLI + WSL + API key auth. Network and the Responses API itself are fine.

What steps can reproduce the bug?

On Ubuntu 24.04.1 LTS in WSL:

1. Install Codex via Homebrew:

``bash
brew install codex
codex --version
# codex-cli 0.57.0
``

1. Verify basic connectivity and API key working from WSL:

```bash
# DNS / TLS checks
ping -c 3 api.openai.com
curl -sS https://api.openai.com/v1/models \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-o /dev/null \
-w "HTTP %{http_code}\n"
# -> HTTP 200

# Responses API check
curl -sS https://api.openai.com/v1/responses \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-4.1-mini",
"input": "Say: hello from WSL."
}'
# -> returns a successful JSON response ("Hello from WSL!")
```

1. Log into Codex using the same API key:

```bash
codex logout || true
rm -f ~/.codex/auth.json

# API key already set in env:
# export OPENAI_API_KEY=...

printenv OPENAI_API_KEY | codex login --with-api-key
# "Successfully logged in"
```

1. Use a minimal local project on the native Linux filesystem (not /mnt/c):

``bash
mkdir -p ~/codex-test
cd ~/codex-test
echo 'print("hello")' > test.py
``

1. Run Codex:

``bash
codex "Say 'hello from codex-cli WSL' and then stop."
``

Observed behavior:

• The TUI starts, but the turn never completes.
• After ~some seconds the CLI complains about timeouts / “stream disconnected”.
• The log file ~/.codex/log/codex-tui.log shows repeated error sending request for url (https://api.openai.com/v1/responses) errors.

Log snippet (timestamps redacted, but this is from the failing run):

INFO event.name="codex.conversation_starts" ... auth_mode="ApiKey" terminal.type=xterm-256color model=gpt-5-codex slug=gpt-5-codex provider_name=OpenAI ...
INFO event.name="codex.user_prompt" ... auth_mode="ApiKey" ... model=gpt-5-codex slug=gpt-5-codex ...
INFO event.name="codex.api_request" ... duration_ms=5002 error.message="error sending request for url (https://api.openai.com/v1/responses)" attempt=0
INFO event.name="codex.api_request" ... duration_ms=5002 error.message="error sending request for url (https://api.openai.com/v1/responses)" attempt=1
INFO event.name="codex.api_request" ... duration_ms=5004 error.message="error sending request for url (https://api.openai.com/v1/responses)" attempt=2
INFO event.name="codex.api_request" ... duration_ms=5002 error.message="error sending request for url (https://api.openai.com/v1/responses)" attempt=3
INFO event.name="codex.api_request" ... duration_ms=5005 error.message="error sending request for url (https://api.openai.com/v1/responses)" attempt=4
WARN stream disconnected - retrying turn (1/5 in 193ms)...
INFO event.name="codex.api_request" ... duration_ms=5006 error.message="error sending request for url (https://api.openai.com/v1/responses)" attempt=0
INFO interrupt received: abort current task, if any
INFO Shutting down Codex instance

During the same time window, a manual curl to https://api.openai.com/v1/responses with the same key and model works fine from the same WSL shell.

What is the expected behavior?

• Codex CLI on WSL should be able to use the API key auth mode successfully, just like:

• Codex CLI on Windows PowerShell does.
• Codex CLI on WSL does when signed in with ChatGPT.
• Specifically, codex should be able to send requests to https://api.openai.com/v1/responses and stream responses without internal “error sending request” failures, given that:

• The same endpoint and key work via curl from the same machine.
• There are no proxies or network restrictions.

Additional information

• ~/.codex/config.toml only contains project trust entries like:

``toml
[projects."/home/jjohnson/codex-test"]
trust_level = "trusted"
``

No custom provider configuration or base URLs are set.

• Environment related to Codex / OpenAI is minimal:

``bash
env | egrep -i 'OPENAI|CODEX|PROXY|HTTPS?_PROXY|ALL_PROXY'
# OPENAI_API_KEY=...
# (no PROXY or CODEX_* variables)
``

• Codex works fine on this same WSL instance when I use “Sign in with ChatGPT” instead of --with-api-key. In that mode, the model responds normally.

• Codex also works fine on the Windows side (PowerShell) with API key auth, against the same project.

So the problem seems localized to codex-cli 0.57.0 → API key / api.openai.com code path on WSL, not to the account, project, key, or network.