Live status
Is the Anthropic API down right now?
The Anthropic API: no problems in our tests in the last 15 minutes · official status (status.claude.com): operational · checked
Updated . While people are reading it, this page is refreshed about once a minute.
Our tests of Anthropic's models
We send "Reply with OK" to Claude Haiku 4.5 every 5 minutes and Claude Fable 5.1 every 30 minutes, and time how long each takes to start answering and to finish. The verdict covers the last 15 minutes: our last three tests.
Last test
- Answered, last 15 min
- 100%
- Answered, last hour
- 100%
- Answered, last 24 h
- 100%
- Median first token, 24 h
- 1.0 s
The last 24 hours
| Model | Last 24 h | First token | Whole reply | Last test |
|---|---|---|---|---|
| Claude Haiku 4.5Every 5 minutes | 43 of 43 answered | 1.0 s | 1.0 s | AnsweredDirect to its maker, |
| Claude Fable 5.1Every 30 minutes | 7 of 7 answered | 1.2 s | 1.2 s | AnsweredThrough OpenRouter, |
No failed tests in the last 24 hours.
What Anthropic's status page says
Anthropic's own status page, read at most once a minute. It reports incidents once Anthropic's team confirms them, so it can lag behind what users and our tests see.
From status.claude.com, read
- Claude Console (platform.claude.com)Operational
- Claude API (api.anthropic.com)Operational
Open incidents
None listed right now.
Resolved on status.claude.com in the last 14 days
Elevated errors for multiple modelsStarted Sep 22, 00:57 UTC · resolved Sep 22, 02:35 UTC
Intermittent error spikes for Claude Mythos 5.1 and Claude Fable 5.1Started Sep 15, 09:48 UTC · resolved Sep 15, 11:14 UTC
What our test covers, and what it doesn't
This page is about Anthropic's API itself, which is what our test calls: the same kind of request your code sends. The Claude apps and Claude Code sit on top of it and have their own parts on status.claude.com.
Each test is one message of a few words, with the reply capped at 32 tokens, tried once, with no retry and no switch to another route. A test counts as failed when the model refuses it as busy (a rate limit or "overloaded"), breaks (a server error), drops the connection, or takes more than 30 seconds. A test our own request broke doesn't count against the Anthropic API.
What to do now
Check the two answers above: if our test and status.claude.com both look fine, the problem is more likely in your request, your key or your rate limits.
Retry with a growing wait between tries (a second, then two, then four), and set a limit on how many.
If only one model fails, send the request to another Claude model for now.
For work that can't wait, ask another company's model.
Ask another model
llmwise has GPT, Gemini, Grok and more in one chat, so an outage at one company doesn't stop your work. The free trial is 5 messages, with a sign-in link by email and no card.
The Anthropic API errors, and how to fix them
The errors people see most with the Anthropic API, what each one usually means, and what to try.
529 overloaded_error
Anthropic's API is too busy to take the request. It isn't your quota.
Retry after a few seconds, doubling the wait each time, and cap the number of tries. If it lasts, try a smaller Claude model, which is often less busy.
429 rate_limit_error
Your organization hit its requests or tokens per minute.
Wait for the time in the retry-after header, then send fewer requests or smaller prompts a minute. Your tier's limits are on the Claude Console's limits page.
500 api_error
Something failed inside Anthropic's API.
Retry once or twice. If many requests fail this way at once, it's on Anthropic's side: check the status above.
413 or 400 invalid_request_error
The request is too big or malformed: a problem with the request, not an outage.
Shorten the prompt or the files you send, and check the request's format. Retrying the same request won't help.
Other AI status pages
Questions
What does a 529 overloaded_error from Anthropic mean?
It's Anthropic's way of saying its servers are too busy for your request right now. It isn't about your account or your limits, and it usually clears within minutes. Retry with a growing wait between tries; if every try fails for a quarter of an hour, check the status here and on status.claude.com.
What's the difference between a 429 and a 529?
A 429 is your organization's own rate limit: too many requests or tokens a minute for your tier. The response's retry-after header says how long to wait. A 529 is Anthropic being busy for everyone. Our tests only ever send one tiny message per model, so a 429 in our results would point at Anthropic's side, not ours.
Does status.claude.com show every API problem?
Not always. Anthropic's status page reports incidents once its team confirms them, which can take a while after errors start, and a small rise in errors may never be posted. Our test sees errors as they happen, but only for the requests we send. Seeing both side by side is the point of this page.
Which endpoint does your test call?
We call Claude from llmwise's servers, the same way llmwise's chat does: directly through Anthropic's API, or through OpenRouter for the models llmwise serves that way. Each result on this page says which route it took.
Claude, GPT, Gemini, DeepSeek, Grok, Kimi, and GLM, in one chat.
See what a message costs before you send it. Free is 5 messages to try; sign in with an email link, no password or card.