Switch to Fallback Models During Recurring Codex Unresponsiveness
OpenAI · Infrastructure & Reliability · · notable
Briefing for: Engineering
What happened
OpenAI’s Codex model has experienced significant unresponsiveness and recurring outages starting March 9, 2026. Despite multiple attempts at mitigation and two 'resolved' statuses, the service has repeatedly reverted to a degraded state with elevated errors.
Why it matters
Any applications, IDE extensions, or automated agents relying on Codex-based endpoints will likely experience timeouts or failed requests. Since the issue has recurred over a 24-hour period, your error handling should account for persistent instability rather than assuming a quick fix.
What this enables
- If you use Codex for production features, temporarily reroute traffic to a GPT-4 variant to maintain service availability.
- If your team's velocity is dropping, check if local IDE plugins or CI/CD scripts are timing out due to these specific service issues.
Get personalized AI briefings for your role at Changecast →