# Anthropic status

> Is the Anthropic API down? Live status, uptime history, and per-region response times for the Claude API, measured since August 2026.

Published: 2026-08-18 | Canonical: https://yoping.me/status/anthropic

## 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.

The live board - the current verdict, 24-hour, 7-day, and 30-day uptime, and per-region response times - is on the HTML page at https://yoping.me/status/anthropic. Those numbers change too fast to repeat honestly in a static mirror, so where an answer below says "the board above", it means that page.

## 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.