TL;DR
OpenAI’s APIs, ChatGPT, and Codex all reported elevated error rates on Saturday, the fourth incident in four days for the company
OpenAI’s status page reported elevated error rates across its APIs, ChatGPT, and Codex on Saturday, marking the fourth service disruption in four days for the company. Users encountered 503 errors with the internal label “biscuit_baker_service_me_circuit_open,” which prevented requests from reaching OpenAI’s servers. The company moved from investigating to monitoring within an hour, saying it had applied a mitigation and was watching for recovery.
The outage hit all three pillars of OpenAI’s product line simultaneously. The API dashboard listed 12 affected components, ChatGPT showed 15, and Codex flagged four, with ChatGPT posting the weakest recent uptime of any OpenAI product over the past 90 days. The timing is awkward for a company that merged ChatGPT, Codex, and its API into a single platform under Greg Brockman in May, precisely so it could concentrate engineering resources rather than spread them thin.
Saturday’s incident was not an isolated stumble, with OpenAI’s own status history showing elevated API error rates on Thursday, elevated errors in Codex Review on the same day, and elevated error rates in both ChatGPT and API image generation on Wednesday. That run follows a pattern that has persisted since autumn 2025, with the status page logging roughly 166 incidents across about nine months, an average of roughly 18 per month. ChatGPT also suffered a major outage on April 20 and again on July 23, just two days before Saturday’s disruption.
The reliability questions land harder now than they would have a year ago. OpenAI’s agentic products passed 10 million weekly users last week, and ChatGPT Work launched this month with the promise that AI agents can finish multi-step tasks across a user’s apps and files, a proposition that depends on the platform staying up. When a chatbot goes offline, users lose a conversation; when an autonomous agent goes offline mid-task, the consequences cascade through the workflows it was managing.
OpenAI is not alone in this struggle. Anthropic’s Claude went down in June, prompting similar questions about whether the infrastructure behind frontier AI models can keep pace with the demands being placed on it. The difference is one of scale and ambition: OpenAI now serves 900 million weekly active users across a unified platform it is preparing to take public later this year, and every outage chips away at the reliability narrative enterprise buyers need before handing real work to an AI agent.
