If your ChatGPT Or Claude crashed this afternoon, you clearly weren’t alone. Monday April 20, 2026, millions of users saw their favorite assistants freeze, display errors or simply refuse to respond.
Between 4:00 p.m. and 5:30 p.m., the reporting curves exploded on DownDetector For ChatGPT And Claudebut also for Google Gemini And Microsoft Copilot. In parallel, Cloudflare recognized degraded performance on several key services, casting doubt on a shared infrastructure failure.
ChatGPT and Claude outage: what happened on April 20, 2026?
The first reports arrived around 4 p.m., with peaks of several thousand complaints for ChatGPT in a few minutes. According to the French press, OpenAI confirmed that an outage was affecting its services and that its teams were working to correct it, while citing a major interruption for its entire ecosystem.
Beyond the website ChatGPTthe API OpenAI and tools like Sora, Dall E, Data Analyst or web browsing were disrupted, although some API integrations appeared to work intermittently. At the house of Anthropic, claude.aiL’API Claude And Claude Code had high error rates, while Gemini And Copilot were especially slow.
Cloudflare, ChatGPT, Claude: why an outage can block everything at once
At the same time, the status page of Cloudflare mentioned degraded performance on Cloudflare Workers, Cloudflare Zero Trust, WARPas well as its analytics and rules tools. This group serves as a CDN and security shield for a huge part of the web, placing itself between the Internet user and the servers of players like OpenAI Or Anthropic.
If this middle layer slows down or blocks requests, upstream sites appear to be down while their own servers are still running, which may explain why the API sometimes responds when the web interface is unusable. However, no actor has confirmed that Cloudflare was solely responsible; other bricks like AWS also count, and a similar breakdown of Cloudflare had already touched ChatGPT And Claude end of 2025.
ChatGPT and Claude restored: how to react during the next outage
According to several French media, the situation generally returned to normal around 5:30 p.m. ChatGPTwith a slightly faster recovery for Claude. For individuals and businesses who rely on these tools on a daily basis, this episode is a reminder of the dependence on a complex infrastructure chain where a single network outage is enough to block production.
In the event of a new breakdown, the good reflex is to first consult the status pages of OpenAI And AnthropicThen DownDetectorin order to distinguish a local bug from a global incident. If the outage is confirmed, it remains possible to temporarily switch to Google Gemini or Microsoft Copilotor to design, in the longer term, tools capable of automatically switching from one model provider to another.