Anthropic's Claude AI models, including Claude Opus 5 and Fable 5, experienced significant elevated errors and service disruptions on August 24, 2026. The outage, which began around 9:50 PM PT (04:50 UTC) and lasted until approximately 00:36 AM PT (07:36 UTC) the following day, resulted in users encountering frequent '529 Overloaded' messages and repeated timeouts.
During the incident, users reported issues across various Claude models, with Opus 5 and Fable 5 being particularly affected. The problems manifested as requests taking longer than usual, leading to multiple retry loops and ultimately failing. While Anthropic's status page initially did not reflect the widespread issues, the company later acknowledged the elevated error rates and began working towards a resolution.
By 08:30 UTC on August 24, Anthropic reported that error rates had stabilized for Opus 5 and Fable 5. The company indicated that efforts were ongoing to fully restore success rates across all affected models. The cause of the widespread errors has not yet been publicly detailed by Anthropic, but the impact was felt by a broad segment of its user base, particularly those relying on Claude for code-related tasks and general conversational AI.
The incident highlighted the critical reliance many developers and businesses place on large language models for their operations. Disruptions of this nature can significantly impact productivity and development cycles, especially for those integrating AI agents into their workflows or using Claude for real-time applications.
Beyond the technical outage, some users also reported idiosyncratic behaviors from Claude during this period, including difficulties in adhering to specific character modes or language nuances. While these instances might be unrelated to the core infrastructure issues, they underscore the complex and sometimes unpredictable nature of advanced AI models, even under normal operating conditions.