Claude Recovers from 80-Minute Outage Spanning Opus 5, Fable, and Mythos Models

Anthropic experienced elevated error rates across several Claude models for roughly 80 minutes on September 22, with impact running from 5:50pm to 7:10pm PT (00:50–02:10 UTC). The problems reached across the company’s product surface, degrading requests to claude.ai, the Claude API, Claude Code, and Claude Cowork. Engineers first flagged the issue as under investigation just before 1:00 UTC and pinpointed the cause about 20 minutes later.

The failures initially concentrated on Claude Mythos 5.1, Claude Fable 5.1, and Claude Opus 5, then broadened to include the 5.0 variants of Fable and Mythos. Recovery was staggered: the Fable and Mythos models returned to normal success rates first, while Claude Opus 5 lagged and was the last to stabilize. Anthropic moved the incident to monitoring at 02:11 UTC once success rates normalized and marked it fully resolved by 02:35 UTC.

Beyond the brief disruption, the incident is a reminder of how much tooling now sits on a single provider’s inference stack—an API hiccup simultaneously knocked out consumer chat, developer CLIs, and agentic workflows. For teams building on Claude, it underscores the value of graceful degradation and retry logic when a shared backend momentarily fails.