OpenAI's API, ChatGPT and Codex go down together — 4th day in a row
One shared dependency took down all three at once this morning — the cost of OpenAI's merged backend, and the case for a fallback behind your agents.
Copy markdown
All three down at once
Starting Saturday morning (~5 a.m. ET), ChatGPT, the developer API and Codex all threw errors together — hung prompts, 503s, dropped logins. OpenAI applied a mitigation and moved from investigating to monitoring within about an hour. If your build broke this morning, it wasn't your code.
Fourth outage in four days
This is the fourth elevated-error incident hitting ChatGPT, the API and Codex in as many days, against roughly 18 incidents a month since autumn 2025. Trailing three-month uptime sits at ~99.7% for ChatGPT and ~99.9% for the API — thin margins if OpenAI is in your critical path.
The unified backend is now the blast radius
The tell was that all three failed simultaneously — a shared dependency, not any single product. OpenAI merged ChatGPT, Codex and the API onto one platform to concentrate engineering; the flip side is that a single circuit-break (503s labeled '…circuit_open') now cascades across everything at once.
What to do: don't single-thread on OpenAI
If OpenAI is load-bearing for you, add a fallback route — a second provider via OpenRouter, or a local model — and wrap calls in retries with backoff so a one-hour blip doesn't become a one-hour outage for your users. Watch status.openai.com for the live incident feed.