If you were in Claude Code on Thursday afternoon European time, you got nothing but errors from about half past three. The status page read “Elevated errors for multiple models”. Affected: Claude Mythos 5.1, Claude Fable 5.1 and Claude Opus 5, which is everything currently at the top of the lineup. Anthropic logs the incident from 13:26 to 16:23 UTC, with impact ending at 16:16 UTC. Just under three hours.
What made it unusual: it wasn’t only Claude. Almost simultaneously, OpenAI reported “Elevated errors across ChatGPT and Codex”, and xAI’s Grok was gone too. A Hacker News thread titled “Ask HN: Why were OpenAI, Claude, and Grok simultaneously down?” collected more than 340 points and over 500 comments, next to three separate threads for ChatGPT, Claude and Grok, all on the front page. For a few hours, a good chunk of the frontier models simply wasn’t reachable.
What’s known about the cause
Anthropic hasn’t explained anything beyond the status entry. OpenAI’s resolution note only says that some Codex Remote Control users may need to pair their phone again. Several tech outlets trace the outage to a failure in Microsoft’s Azure East US region, and an OpenAI spokesperson reportedly described a routing error starting at 7:43 am Pacific. The Azure connection hasn’t been officially confirmed.
Hacker News did what Hacker News does: Cloudflare was the first suspect, then Downdetector’s methodology got taken apart. One commenter summed up the timeline dryly: ChatGPT and Grok were back, Claude wasn’t. “Typical.” Anyone who’s watched the Claude status page lately knows what he means. The history lists 25 incidents in August alone.
What people did in the meantime
The more entertaining part of the thread is the workarounds. One person used the downtime to try Sonnet 5 again and found it “faster and less vocal but not much dumber”. Another described how his work has changed: during GitHub outages he used to just code something else. During model outages, what’s left is catching up on email and thinking about architecture. And at least one commenter recommended a coding harness that can switch models through OpenRouter at any moment.
That’s the point for me. Three providers, one outage, one afternoon. If your work runs entirely on a single model, a day like this leaves you with nothing to do. That applies to me as much as to any company pushing agents into production right now. A fallback model in your config isn’t paranoia anymore, it’s basic craft. That the outage hit on GPT-6 Astra’s launch day probably didn’t thrill OpenAI. And Anthropic, which shipped Claude Code 2.1.260 the same day, likely wasn’t happy either.
Sources: Claude Status: Elevated errors for multiple models, OpenAI Status: Incident History, Hacker News: Ask HN: Why were OpenAI, Claude, and Grok simultaneously down?, Hacker News: Claude outage – Resolved