Anthropic Resolves Errors Across Claude Opus 5, Mythos 5.1 and Fable 5.1
The roughly 80-minute incident affected requests to three Claude model lines before success rates returned to normal; Anthropic did not publish a root-cause explanation.
Edited by Tyronne Panaino
Anthropic resolved a September 22 incident that produced elevated errors on requests to Claude Opus 5, Claude Mythos 5.1 and Claude Fable 5.1. The company's official status page records an impact window from 00:50 to 02:10 UTC, making the disruption about 80 minutes long, and marked the incident resolved at 02:35 UTC.
The status page showed all listed Claude services as operational when this article was reviewed. That is the clearest current signal for users and teams depending on those model endpoints, but the record does not quantify how many requests failed, how broadly customers were affected or which technical fault caused the errors.
Recovery happened in stages
Anthropic began investigating at 00:57 UTC and reported at 01:17 UTC that it had identified the cause. That update named requests to Mythos 5.1, Fable 5.1 and Opus 5 as affected, but it did not disclose the cause itself.
At 01:35 UTC, Anthropic said success rates for Fable 5 and 5.1 and Mythos 5 and 5.1 had returned to normal while work continued on Opus 5. The company then reported at 02:11 UTC that success rates had returned to normal across the affected models and moved the incident into monitoring. Its final entry closed the impact period at 02:10 UTC and marked the issue resolved.
That sequence matters for operators reviewing failed jobs or degraded agent runs. Recovery was not presented as a single simultaneous switch: the Fable and Mythos families were reported healthy before Opus 5. Teams with workflows spanning several models should therefore compare their own failure timestamps with the individual recovery updates instead of treating the final resolution time as the only useful checkpoint.
What the record establishes
The official incident log establishes the affected model names, the investigation and recovery sequence, the recorded impact interval and the final resolved state. It also shows the broader Claude status dashboard reporting claude.ai, Claude Console, the Claude API, Claude Code, Claude Cowork and Claude for Government as operational at review time.
The same page does not say whether the incident was tied to one region, one request interface or a shared model-serving dependency. It provides no failure-rate graph, customer-impact count, remediation detail or post-incident analysis. Those gaps limit what can responsibly be concluded about the severity and recurrence risk. A future technical review from Anthropic would be the next verifiable source for the root cause and any preventive changes.
Practical implications for Claude users
Users who saw errors during the recorded window can now distinguish this provider-side incident from a continuing outage. Reliability teams can correlate retries, queue delays and failed evaluations against the 00:50-to-02:10 UTC interval, while treating any problems after the reported recovery as a separate issue that needs fresh diagnosis.
The incident also illustrates why model-level updates matter even when a dashboard later returns to green. A final operational state confirms service restoration; it does not reconstruct which model recovered first or explain the failure. The timestamped update trail supplies that narrower operational history without supporting broader claims about Anthropic's infrastructure.
Status
Confirmed. Anthropic's official status record says the multi-model error incident is resolved and all listed systems were operational at review time. Internal confidence is medium because the evidence comes from one company-controlled operational source and no independent incident analysis or root-cause report was available.
Sources
Update note: Last reviewed September 22, 2026. We will revise this post if Anthropic publishes a root-cause analysis or materially changes the incident record.
Sources
- Anthropic Claude status page — official
Drafted with AI assistance from source briefs; reviewed for citation completeness and label accuracy.