Anthropic's Claude platform suffered elevated error rates across multiple flagship models early Tuesday, with services disrupted for roughly 90 minutes before the company resolved the incident. According to the company's official status page, the outage impacted Opus 4.8, Opus 4.7, Opus 4.6, and Sonnet 4.6 โ€” a significant chunk of their production model lineup. The affected services included claude.ai, the Claude API at api.anthropic.com, Claude Code, and the newer Claude Cowork product.

What Happened

The incident began around 00:37 UTC on June 22, 2026, when Anthropic's monitoring systems detected elevated error rates across several models. The company immediately opened an investigation, with engineers working to identify the root cause. Within about 34 minutes โ€” at 01:11 UTC โ€” Anthropic had identified the issue and was actively implementing a fix.

Recovery Timeline

The restoration process unfolded in stages rather than all at once. Opus 4.8 came back online first at 01:16 UTC, followed by Haiku 4.5 at 01:33 UTC. Opus 4.7 recovered around 01:56 UTC, with Sonnet 4.6 following shortly after as the team monitored results. By 02:06 UTC โ€” approximately an hour and a half after initial detection โ€” Anthropic declared the incident resolved and stated they were monitoring for any residual issues.

Services Affected

The outage wasn't limited to a single interface. Users of claude.ai's web chat, developers hitting the API endpoint api.anthropic.com, those using Claude Code for CLI-based development, and Cowork users all reported problems during the window. For teams building production systems on Anthropic's infrastructure, this meant failed requests and potential cascading issues in dependent applications โ€” exactly the kind of thing that makes on-call engineers wince.

Key Takeaways

  • Multiple high-tier models went down simultaneously, suggesting a shared infrastructure or deployment issue rather than model-specific problems
  • Total incident duration was roughly 90 minutes from first detection to resolution
  • The staggered recovery suggests careful validation steps were taken before marking each model healthy
  • Claude Cowork โ€” Anthropic's collaborative workspace product โ€” was among those affected, indicating the outage had broad reach across their product portfolio

The Bottom Line

This one's a reminder that even the most heavily-funded AI labs deal with operational turbulence. The fact that Anthropic resolved it in under two hours and communicated status updates throughout is solid incident management โ€” but for teams running production workloads, incidents like this underscore why you want fallback options and proper retry logic in place when your AI provider hiccups.