- Sources: incident zftg3gqkmv18, incident history feed, Anthropic incident 18:40 UTC, Anthropic incident 21:34 UTC, HN 49056194
- Summary: The most recent incident, titled elevated errors for Opus 5, opened as investigating at 2026-07-26 09:17 UTC, was identified at 09:45 UTC, moved to monitoring at 10:34 UTC, and was marked resolved at 10:44 UTC. The incident history feed lists model-serving error incidents on each day from 2026-07-21 through 2026-07-26. The heaviest day was 2026-07-25 with three separate entries. Two of them are rated major: one ran 18:40 to 19:44 UTC with elevated errors on Mythos 5, Fable 5, Opus 5, and Haiku 4.5, listing claude.ai, the Claude API, Claude Code, and Claude Cowork as affected components, and was marked identified within four minutes. The second ran 21:34 to 22:08 UTC with elevated errors on Fable 5, Sonnet 5, Haiku 4.5, and other models across the same components. A third, minor entry earlier that day covered about 10 minutes of elevated Sonnet 4.6 and Sonnet 5 errors. Two of the earlier entries in the six-day run were broader service disruptions affecting document creation in claude.ai, Cowork Remote, Claude Code, Claude Code on the Web, Claude Tag, and Claude Design. No root cause is published for any of them.
- Why it matters: Six consecutive days of model-serving errors is a base rate, and retry budgets and fallback routing on a production request path are worth sizing against that rather than against a single incident.
- Follow-up: Watch for a published cause for the 2026-07-21 to 2026-07-26 cluster and for whether the daily cadence continues.
send feedback on this story