Anthropic opened an investigation into elevated errors on five Claude models on October 25, 2026, with the incident posted at 06:14 UTC in Investigating status. The affected models are Claude Mythos 5.1, Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 5.5. For business users this matters because the same event touches claude. ai, the Claude API, Claude Code and Claude Cowork at once.
Errors across five models and four services
The incident titled "Elevated errors for multiple models" lists four affected services: claude. ai, Claude API at api. anthropic. com, Claude Code and Claude Cowork. Anthropic said it was investigating elevated errors on requests to the five models and would provide an update as soon as possible. Seven minutes after posting, the record remained in Investigating status with no further updates published. The Claude Status page listed it as the unresolved incident for October 25, 2026.
At the time of posting, four components were marked as Partial Outage: claude. ai, Claude API, Claude Code and Claude Cowork. Two other components stayed Operational: Claude Console at platform. claude. com and Claude for Government. The 90-day uptime figures shown were 99.58 percent for claude. ai, 99.64 percent for Claude API, 99.58 percent for Claude Code and 99.57 percent for Claude Cowork. Claude Console showed 99.84 percent and Claude for Government showed 100 percent.
The October 25 event follows three resolved incidents on October 24. "Elevated errors for Claude Fable 5 and Claude Fable 5.1" was resolved at 05:53 UTC, with impact from 05:00 to 05:35 UTC; investigation began at 05:16, cause identified by 05:24, fix deployed and monitoring by 05:41. "Elevated errors for Claude Opus 5.5" was resolved at 00:37 UTC after investigation from 00:31, with impact from 00:23 to 00:34 UTC. A third incident for Fable 5.1, Sonnet 5.5 and Haiku 5.5 was resolved at 00:11 UTC, with impact from 23:28 to 23:59 UTC on October 23.
What repeated outages mean for operations
Companies that run support, sales or internal workflows on claude. ai, the API, Code or Cowork face a single point of sensitivity when several models fail together. A team using Mythos 5.1 for drafting and Haiku 5.5 for high-volume classification could see both paths degrade at the same time. Small firms without fallback models feel this as paused work, while large firms with API routing can shift traffic if Console access and monitoring stay available.
The record leaves open what triggers the errors, how long resolution takes and whether fast mode requests play a role again. An Identified update on the October 23 late incident cited fast mode requests to Opus 5.5, and a Monitoring update at 23:54 UTC said success rates were returning to normal. This news alone does not mean core data was lost or that Console usage reporting was affected. Before choosing a model, ask the vendor which request types were affected, what success-rate threshold defines recovery, and where status updates appear first.
The marker to watch is the next status change for the October 25 incident from Investigating to Identified, Monitoring and Resolved, with stated impact times. A short window similar to the 11-minute and 31-minute impacts on October 24 would point to contained instability. Repeated multi-model entries across several days would signal a need for formal retry and failover rules.
