Is Replicate down right now?

No. Our checks are finding no problems with Replicate right now. That verdict is live, and the banner at the top of this page carries it too. We re-check Replicate continuously using its official status feed, direct network probes and reports from visitors, stopping at the first check that gives a clear answer, and the page updates itself as soon as anything changes. A green verdict does not rule out a fault affecting only some users or regions.

The problems visitors report most often on this page are API errors, no response / timing out, 5xx error pages, rate limited, high latency, can't connect and slow or laggy. Those are the options the report button offers for Replicate, chosen for how this service actually fails rather than a generic list, and the "user reports" chart above breaks down which ones people are picking right now.

Our checks run from one place on the internet, so a fault limited to your region or network can be real while our probes succeed. With Replicate specifically, individual model servers or GPU clusters can fail independently, regional API endpoints may have different availability, and rate limits apply per account rather than globally. That is why this page reports what it observed rather than declaring the service healthy, and why reports from other visitors count toward the verdict.

For Replicate, run a simple curl request to the Replicate API health endpoint or submit a test inference job to a small model and check the response code. A problem that follows you across networks and devices is very unlikely to be yours.

If this page shows evidence of problems, the fault is most likely on Replicate's side. Specifically: do not retry requests in tight loops, as this consumes your rate limit quota and delays recovery; instead, implement exponential backoff and check the status page before retrying. Reporting it here helps other people see they are not alone.

Continuously, around the clock. Replicate publishes a machine-readable status feed, so our first check reads it directly and reports what Replicate itself says. If that feed is unreachable we fall back to a direct HTTP probe, then an ICMP ping, then reports from visitors. The verification tier box on this page names whichever check produced the verdict you are reading, so you can see how direct the evidence is.

Yes, at status.replicate.com, and it is linked further up this page. That is Replicate's own account of their service and it is the place to look for their estimate of a fix. We read it where we can, but we publish our own observations beside it rather than only repeating theirs, because an operator's status page can lag a problem their users are already hitting.

No. status.buzz is an independent monitoring site operated by LuzLab LLC, with no connection to Replicate or replicate.com. It reads publicly available endpoints and status feeds, and publishes what it observed. For account or billing problems you need Replicate's own support, not us.