Anthropic's services failed worldwide on the evening of 29 July. Elevated errors across all models began at 19:49 UTC, affecting claude.ai, the Claude API and Claude Code at once. Full resolution came at 22:36 UTC, just under three hours later.
What users saw
Requests failed with 529 Overloaded — the status code a service returns when it cannot take the load, rather than when something has broken. That distinction matters for interpretation, though Anthropic has not confirmed capacity as the cause.
The timeline
Anthropic began investigating at 19:49, said it had identified the issue at 20:33, and did not publish a cause or a recovery estimate during the incident. A spokesperson confirmed afterwards that "service has been fully restored across Claude.ai, Claude Code, and the Claude API." No post-incident explanation has followed.
Why the API half matters more
A chatbot outage is an inconvenience. An API outage stops production systems, and Claude Code failing takes engineering teams offline mid-task rather than merely mid-conversation. As frontier models move from assistants to dependencies inside deployed software, single-provider downtime stops being a consumer story.
The silence
No root-cause analysis has been published. For a company selling into enterprises that increasingly route production workloads through a single API, the absence of a postmortem is itself the notable part — buyers negotiating uptime terms have nothing to read.