Update: OpenAI's API and Codex hit a 4th straight day of outages
Saturday's incident hit ChatGPT, Codex and the image and agent APIs; the status page flashed green while requests hung. It's the pattern, not one bad day.
Copy markdown
What's actually down
Saturday's incident opened ~5:11am ET, hitting ChatGPT, the API, Codex, gpt-image-2 and the Workspace Agent — hung prompts, dropped logins, failed uploads and 503s. It's intermittent, not a clean hard-down, so retries sometimes work and sometimes don't.
Don't trust the status page
OpenAI's dashboard flickered to 'operational' mid-incident while requests kept failing — a sign of a shared backend dependency, not isolated services. Wire your own canary (a tiny completion every few minutes) instead of polling status.openai.com.
It's four days, not one
Elevated-error incidents have recurred across ChatGPT, the API and Codex since July 22–23, several needing mitigations before recovery. Three-month ChatGPT availability sits near 99.7% vs the API's 99.9% — thin margin if uptime is load-bearing for you.
Anthropic just shipped a fallback lever
Timely: the Claude API's new fallbacks:"default" mode (beta header server-side-fallback-2026-07-01) auto-routes to recommended models server-side. It targets refusals, not outages — but it's the same instinct: stop hard-coding one model as your only path.