- Sources: incident mfdtrknpxghq, incident lhqp09kxq7pb, incident rkk5x44tndw9, status history feed, HN 49066591, HN 49068029
- Summary: Anthropic's status page opened three separate model-error incidents on 2026-07-27. The first, titled Elevated errors on Claude Opus 5, was under investigation at 08:16 UTC and resolved at 09:05 UTC, with errors reported back to baseline as of 09:03 UTC. The second, same title, was under investigation at 11:27 UTC and resolved at 12:30 UTC, with errors back to baseline as of 11:47 UTC. The third, titled Elevated errors on Claude Opus 5 and Haiku 4.5, was under investigation at 13:39 UTC, noted at 14:04 UTC that Haiku 4.5 was also returning errors, and resolved at 14:36 UTC with errors back to baseline as of 14:34 UTC. None carries a cause. Reading the status history feed, fourteen model-error incidents were opened between 2026-07-21 and 2026-07-27: two on 07-21, three on 07-22 plus one resolving 07-23, one on 07-24, three on 07-25, one on 07-26, and three on 07-27. Two further service-disruption incidents on 07-21 and 07-22 are a different failure and are counted separately. No incident in that range publishes a root cause. The count is bounded at 07-21 because that is where the feed read at this run ends, so it is a floor rather than the full span of the pattern.
- Why it matters: Teams putting Claude models on a request path are absorbing a repeated failure mode at roughly two incidents a day with no published explanation, so retry, timeout, and model-fallback behaviour is load-bearing and there is no vendor statement to size it against.
- Follow-up: Watch for a root-cause note or an incident summary covering the 2026-07-21 to 2026-07-27 run, and for whether the rate continues.
send feedback on this story