Update: Claude outage over — API, Code, claude.ai back to baseline
Anthropic cleared the cross-model incident overnight — error rates back to baseline. The timeline, the load-driven cause, and how to harden for next time.
Copy markdown
The all-clear
Anthropic marked the incident resolved at 22:36 UTC on Jul 29 (2:36 AM Dubai); success rates recovered across all models and status.claude.com now reads "All Systems Operational." If Claude Code or your API calls were erroring earlier tonight, that's why — and it's over.
What it hit — and for how long
Elevated errors spanned the API (api.anthropic.com), Claude Code, and claude.ai, with Opus 5 and Haiku 4.5 among the affected models. Anthropic opened the investigation at 19:49 UTC and saw partial recovery by ~21:38 UTC, so user-facing errors ran on the order of two hours — not a full day.
The cause — and how not to get burned next time
The status page points to elevated requests and latency to Claude models — load, not a bad deploy. If your app calls Claude directly, add exponential-backoff retries on 429/529 responses and wire a fallback model or provider so a single-lab blip degrades gracefully instead of taking your product down with it.