Fix request timeouts

Timeouts show up as missing first bytes, client timeout / response header timeout errors, or eventual 503 cooldown responses. They are not the same as bad tokens.

Likely causes

1. Long reasoning or brief upstream congestion — see cooldowns & retries.
2. Local network / proxy — try .com vs .net; inspect corporate proxies and TLS-intercepting tools.
3. Client timeout too aggressive — raise the client limit for long jobs and use bounded backoff.
4. Too much parallelism — reduce concurrent Codex / Claude Code / batch jobs to 1–3.

Steps

1. Retry once on the other relay host (tokens are shared).
2. Reproduce with a tiny prompt on the same model.
3. Check Status for maintenance.
4. If you see no_account_available / cooling down, wait—do not hammer retry.
5. After ~15 minutes, contact support@4aicode.com with time, model, tool, and trace_id (never the token).

401 / 429 vs timeout

Related

• Performance
• Models
• Codex setup