ChatGPT recovers after daylong global outage

If you woke up on Tuesday and couldn't get ChatGPT to do your job, you definitely weren't alone.

OpenAI suffered a massive, daylong outage on Tuesday that knocked out ChatGPT, its text-to-video model Sora, and the core APIs that power hundreds of external apps.

A rough Tuesday for AI workers

According to Downdetector, the issues kicked off around 3:00 AM ET, quickly spreading across multiple regions worldwide.

Users trying to prompt ChatGPT were greeted with a generic error: "Hmm…something seems to have gone wrong." Across Reddit and X, people reported that the platform was completely inaccessible or agonizingly slow to respond, though a handful of lucky users managed to slip through unscathed.

OpenAI's status page initially acknowledged "elevated error rates and latency" across ChatGPT, Sora, and its APIs. By 5:23 AM ET, the company added a specific incident note for Sora before downgrading both systems to a "partial outage."

The fallout didn't stop at OpenAI's front door. Perplexity, the AI search engine that relies on OpenAI models for part of its pipeline, was dragged down too. Perplexity's own status page reported slowness and elevated errors starting around 7:00 AM ET.

By 6:32 PM ET, OpenAI reported that its APIs had staged a full recovery and that nearly all ChatGPT features were back up and running. The only stubborn holdout was voice mode, which OpenAI noted was still seeing elevated error rates.

Why it matters

We think this daylong outage exposes just how fragile our collective reliance on a single AI provider has become.

For knowledge workers and developers, ChatGPT isn't an experimental novelty anymore—it is everyday infrastructure. When OpenAI goes dark, productivity hits a wall. More importantly, third-party apps like Perplexity getting caught in the crossfire shows how vulnerable downstream companies are when they rely heavily on one dominant API.

The fact that ChatGPT, Sora, and enterprise APIs degraded together suggests their underlying serving infrastructure remains tightly coupled. In our view, if you are a developer running business-critical software on large language models, this is a clear wake-up call to set up multi-model fallbacks. Every hour of downtime makes Anthropic, Google, and open-source alternatives look much more attractive.