Claude Went Down for Three Hours Again: Multiple Models Erroring on August 24, and Claude Code Users Hit 529

Anthropic's status page reported elevated errors across multiple models at 05:06 UTC on August 24, marked the cause identified 21 minutes later, and was back to normal around 08:30 UTC — roughly three hours end to end. Mythos 5, Fable 5, Opus 5 and Opus 4.8 were affected, with Claude.ai, the API, Claude Code and Cowork all listed as partial outages while the Console and Claude for Government stayed up. Pages loaded but sending a message failed; Claude Code surfaced it as 529 Overloaded, and one Reddit user reported a nine-hour run cut short. Anthropic has not disclosed a root cause publicly. This wasn't a one-off: model-affecting incidents were also logged on August 6, 7, 14, 16, 18, 19 and 20.

The Shape of This One

It wasn't the kind of outage where the service disappears. Web, mobile and desktop all loaded, the input box was there, and the error arrived when you hit send — a few refreshes for someone chatting, and a stretch of consecutive failures for anyone with requests inside a script. Claude Code users saw 529 Overloaded, and long tasks died mid-run with nothing to resume from. Judging by the status page cadence, Anthropic moved quickly: investigating at 05:06, identified at 05:27, recovery over the following stretch, all green around 08:30. What hasn't come is a root cause — the public record stops at "identified and fixed," which isn't enough for a team that has to write its own incident review.

The Frequency Matters More Than This Instance

A single three-hour incident isn't remarkable. Where it sits in the month is. August 6 (the more severe one), 7, 14, 16, 18, 19 and 20 all carry incidents touching Opus 5, Haiku 4.5, Fable 5 or Mythos 5 — mostly in the form of elevated error rates on some models or degraded performance. Third-party monitor StatusGator has logged over a hundred Claude events this year; that count includes minor degradations and is broader than Anthropic's own incident list, so it isn't "a hundred outages," but it's fine for reading the trend. Frontier capacity is being consumed by agentic load fast. The same conversation that a human measures in seconds, an automated pipeline measures in hours, and the peak shapes are nothing alike. The 529 code says "overloaded" on the tin — it's a capacity signal, not network flakiness.

Design for Availability Instead of Assuming It

If you only use the model in a browser, the story ends here. If you have it inside CI, a cron job or a production path, three things are worth doing now. Put exponential backoff on every call and handle 429/529 separately from genuine business errors. Checkpoint long tasks so a run resumes from where it stopped rather than starting over. And keep a cross-vendor fallback path — not necessarily active-active, but enough that "this provider is down today" doesn't mean "this pipeline is down today." Availability commitments generally live in enterprise contracts; individuals and small teams get best-effort. Given that, treating interruptions as a recurring normal event you design around is less work than refreshing the status page each time.

via: Anthropic's status page, Android Authority, Notebookcheck