Anthropic is managing its eighth consecutive day of elevated error rates, with an incident on August 20 touching nearly every Claude product at once.
The status notice, filed at 19:16 UTC, flags claude.ai, the Claude API, Claude Code, and Claude Cowork as partially down. The Claude Console and Claude for Government kept working, which points the finger at the model-serving layer rather than accounts or administration.
Anthropic’s update says only that some Claude models are seeing elevated errors. It has not said which models, why, or when the trouble will clear.
The August 20 event is the third disruption in about a day. A Google connectors issue was resolved earlier that evening, and on August 19 the company fixed errors on Opus 5 and Haiku 4.5. Earlier in the week, Mythos 5, Fable 5, Sonnet 5, and other models degraded in sequence.
Agentic products make the impact sharper. Claude Code and Cowork run long multi-step sessions, so an elevated error rate kills agent runs mid-task instead of merely failing a chat message. Teams that have routed CI pipelines or internal tooling through the API feel the outage on the most load-bearing line.
The stretch is precisely why multi-model routing services like Ramp’s Router exist, as customers weigh the risk of leaning on a single vendor whose bad week becomes their downtime.
