Anthropic status

How we check Anthropic

We watch api.anthropic.com from all five of our probe regions, every 2 minutes. A region that sees a failure does not make this page say "down" on its own: a second region has to agree first, the same confirmation rule every YoPingMe monitor uses.

We pin the exact response code the API returns to a request with no valid key, rather than assuming any 2xx. An unauthenticated request to a modern API is normally rejected on purpose, and treating that rejection as an outage would report every healthy day as one.

What is Anthropic?

Anthropic's API is how most products that use Claude actually talk to it: a request carrying a prompt goes out, a completion streams back, and the product built on top never has to host or run the model itself. For a growing set of tools, from coding assistants to customer-support systems to internal automation, that API call sits directly in the critical path between a user's action and the response they see.

An Anthropic incident tends to show up differently than a typical web outage. Because most calls happen server-side, from inside another product's backend, a degradation rarely looks like a page failing to load. It looks like a request that times out, comes back with an error, or streams noticeably slower than usual. In a chat-style interface, that often means a visible pause before text starts appearing, which reads as "broken" to a user even when the request technically succeeds a few seconds later than expected. In a background job, calls quietly piling up against a slow API can back up a queue well before anyone watching dashboards notices anything is wrong.

Response time matters almost as much as plain availability here. A product built around a fast, conversational back-and-forth can feel completely broken during a period of elevated latency, even if every request the API receives is still eventually answered correctly. That is the case for reading the response-time figures above alongside, not instead of, the up-or-down state: a service can be technically "up" and still be unusable for anything time-sensitive.

Incidents at this scale are typically partial: a specific model or a specific region degraded rather than every request across the board failing at once. If a Claude-powered feature in your own product starts erroring and the board above shows Anthropic healthy, the more likely culprits are an expired or rotated API key, a rate limit you have hit, or a request payload that has drifted from what the current API version expects.

Frequently asked questions

Is the Anthropic API down right now?

Check the board above; it reflects our own probes against the API, refreshed every couple of minutes from all five of our regions.

My application using Claude is erroring out. Is that an Anthropic outage?

Not necessarily. An expired API key, a rate limit, or a request that no longer matches the current API's expected shape produces the same kind of error a real outage would. If the board above shows Anthropic healthy, check your own account dashboard and error message first.

Do API incidents affect Claude.ai the consumer product too?

Not always. Claude.ai and the developer API share infrastructure but are tracked and can degrade separately, so an API-only incident can leave the consumer app working fine and vice versa. This page watches the API specifically, since that is the dependency most integrations sit on top of.

What does a partial Anthropic incident usually look like?

More often elevated latency or errors on a specific model or a specific region than a full stop of every request. That is why the per-region response times above are worth reading alongside the plain up-or-down verdict, not instead of it.