Anthropic began investigating elevated error rates at 00:57 UTC on September 22, initially naming Claude Mythos 5.1, Claude Fable 5.1 and Claude Opus 5 among the affected models. By 01:17, the company said it had identified the cause and was working on a fix.

Recovery came in stages. At 01:35, Anthropic reported normal success rates for Fable 5 and 5.1 as well as Mythos 5 and 5.1, while Opus 5 was still experiencing problems. Success rates across the affected models had returned to normal by 02:11, and the incident was formally marked resolved at 02:35.

The actual impact lasted roughly 80 minutes

Anthropic’s status record separates the incident updates from the period when users were actually affected. It places the impact between 00:50 and 02:10 UTC, giving the disruption a duration of about one hour and twenty minutes.

At the time of verification, Anthropic’s status page listed claude.ai, the Claude API, Claude Code, Claude Cowork and its other main services as operational. No new incident was recorded for September 23.

The missing piece is the root cause. Anthropic publicly stated that it had identified it, but did not describe the failing component, an infrastructure change, an external dependency or any other technical trigger. There is currently no solid basis for assigning the incident to a particular deployment or workload.

It follows several other September reliability issues

September has not been completely quiet for Claude. Anthropic reported intermittent error spikes affecting Mythos 5.1 and Fable 5.1 on September 15, another elevated-error incident involving those models on September 11, and increased API latency for some customers on September 10.

The September 22 incident was relatively short by itself. Repeated failures still matter more for companies running Claude inside automated systems than they do for someone occasionally opening the chatbot. API users need retry logic, sensible timeouts and, for critical workloads, a fallback path rather than assuming permanent availability.

It happened on the same date as Opus 5.5, but no connection has been established

September 22 was also the date Anthropic announced Claude Opus 5.5. The company says its new model is cheaper to operate than Opus 5 and ships with stronger safeguards for areas including cybersecurity and biology.

That timing is notable, but it is not evidence of causation. Anthropic’s incident updates specifically referred to models including Opus 5, Fable 5.1 and Mythos 5.1, and the company has not connected the outage to the Opus 5.5 rollout.

So the latest publicly documented operational problem at Anthropic is, for now, a resolved multi-model reliability incident with a known impact window and an undisclosed technical cause.