openclaw - 💡(How to fix) Fix [Bug]: OpenAI gpt-5.4-mini/nano times out at ~130s after upgrade — model catalog strips api field, falls back to openai-responses transport

Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

After upgrading from 2026.4.23 to 2026.4.29, all requests to openai/gpt-5.4-mini and openai/gpt-5.4-nano time out at ~130 seconds; downgrading to 2026.4.23 immediately restores normal behavior.

Steps to reproduce

1. Configure openclaw with openai provider using a direct API key and gpt-5.4-mini as primary model.
2. Upgrade from 2026.4.23 to 2026.4.29 (npm install -g [email protected]).
3. Restart gateway: openclaw gateway restart.
4. Send any message to the agent.
5. Observe: request hangs for ~130 seconds, then times out with no response.

Expected behavior

In 2026.4.23, the same configuration routes requests to Chat Completions (openai-completions transport) and responds within a few seconds.

Actual behavior

Requests hang for ~130 seconds and time out. Based on observed behavior, the model catalog appears to strip the transport/api field resolution, causing models to fall back to openai-responses transport (Responses API) which does not complete the request.

OpenClaw version

2026.4.29 (regression from 2026.4.23)

Operating system

macOS 15.4

Install method

npm global

Model

openai/gpt-5.4-mini, openai/gpt-5.4-nano

Provider / routing chain

openclaw -> api.openai.com/v1 (direct API key)

Additional provider/model setup details

Auth: api_key mode via env.vars (OPENAI_API_KEY). No Codex OAuth involved. Provider defined in agents/dev/agent/models.json under providers.openai with baseUrl: https://api.openai.com/v1. Primary model set in openclaw.json under agents.defaults.model.primary. Direct curl to https://api.openai.com/v1/chat/completions with the same API key returns a valid response in <2s, confirming the issue is internal to OpenClaw's routing layer.

Logs, screenshots, and evidence

- Direct curl to Chat Completions API: responds correctly in <2s
- Same model via openclaw 2026.4.29: hangs ~130s, no response
- Downgrade to 2026.4.23: responds correctly in <2s
- No gateway error logs captured before timeout

Impact and severity

Affected: Any user running openai provider with direct API key on 2026.4.29 Severity: Critical — completely blocks agent usage Frequency: 100% reproducible (every request, every session) Consequence: Agent is entirely non-functional on 2026.4.29 with this setup

Additional information

Last known good: 2026.4.23 First known bad: 2026.4.29 Workaround: downgrade to 2026.4.23 (npm install -g [email protected]) This is distinct from issues #38706 and #41282 which affect Codex OAuth paths — this bug affects standard direct API key setups with no Codex involvement.