Anthropic resolves ~100-minute outage that hit error rates across all Claude models

Anthropic’s Claude experienced elevated error rates and latency across every model tier during an incident on July 29, 2026. The disruption ran from roughly 19:45 to 21:26 UTC — about an hour and forty minutes — before success rates recovered. Engineers first flagged the problem, worked through a period of degraded requests and slow responses, then confirmed recovery and closed the incident while continuing to watch for regressions.

The status page offers no root-cause detail, so the underlying trigger is unknown. What’s notable is the blast radius: the fault degraded all models simultaneously rather than a single endpoint, pointing to a shared infrastructure or routing layer rather than a model-specific fault.

For teams that build on the Claude API, an outage of this kind is a reminder that a single upstream provider is a hard dependency. Even a sub-two-hour window of elevated errors can cascade through production systems that lack retry logic, graceful degradation, or failover to an alternate model or provider.