How often is Claude down?
Over the last 90 days, Anthropic declared 40 incidents, 13 of them major or critical. Mean time to recovery is 2 h 51. These figures come from Anthropic's own status page, which publishes incidents but not the aggregate.
| Month | Incidents | Major / critical | Mean recovery |
|---|---|---|---|
| 2026-03 | 0 | 0 | — |
| 2026-04 | 0 | 0 | — |
| 2026-05 | 0 | 0 | — |
| 2026-06 | 0 | 0 | — |
| 2026-07 | 40 | 13 | 2 h 51 |
| 2026-08 | 0 | 0 | — |
How these numbers are computed
An incident is counted in the month it started, in UTC, so an incident spanning two months is never counted twice. Recovery time is the gap between Anthropic opening the incident and marking it resolved, using their own timestamps.
What these numbers are not: the cumulative "major-incident time" (24 h 55 over 90 days) is not total unavailability. Several of those incidents were partial — one degraded component, not the whole service — and an incident held open for a long time does not mean nothing worked throughout.
Longest incidents
| Elevated errors across many models | 2026-07-30 | 4 h 51 |
| Elevated errors for multiple models | 2026-07-16 | 4 h 16 |
| Elevated errors on Haiku 4.5 | 2026-07-21 | 3 h 34 |
| Elevated errors across all models | 2026-07-29 | 2 h 47 |
| Elevated errors on several models | 2026-07-21 | 2 h 44 |
Source: status.claude.com — read on 8/2/2026, 8:53:43 PM UTC. Live status →