Is Replicate down?
Loading the latest status…
Is Replicate down right now? Live status from Replicate’s status page, our own checks and user reports, updated every minute.
- Official status
- Reachable
- Reports in the last hour
Detailed data for this service is briefly unavailable. The status above comes from our latest check.
Replicate’s status page
- Replicate
Not working for me
Pick what’s broken. Your report shows up on the chart below within a few minutes.
Last 24 hours
- Operational
- Degraded
- Partial outage
- Major outage
- No data
- Reachability check (api.replicate.com)
By region
| Region | Replicate API api.replicate.com |
|---|---|
| N. Virginia | |
| Oregon | |
| Frankfurt | |
| Singapore | |
| Tokyo | |
| São Paulo |
Each row shows the latest result from that region. How we check
Last 90 days
We have checked Replicate ourselves since Oct 28, 2026. Earlier days come from its status page.
1 day with problems in the last 90 days.
Recent incidents
Incidents: 1 in the last 30 days and 1 in the last 90, lasting a median of 31 min.
- Prediction issue on OpenAI models via ReplicateDegraded31 min
If Replicate isn’t working for you
Replicate is a platform for running and deploying machine learning models, from image and video generators to language models, through an API.
Common Replicate errors
| Error | What it means | Usually | What to do |
|---|---|---|---|
Request was throttled. Expected available in 1 second. | A 429 error: you hit Replicate's API rate limit. | Your side | Wait the time given and retry with backoff. Source |
E1000 - Unknown Error | An uncategorized error, which can come from temporary system issues or network problems. | An outage | Retry the prediction and check your connection; contact support if it continues. Source |
E1001 - Model Out of Memory (OOM Killed) | The model ran out of memory, usually because the input is too large. | Your side | Use smaller images or shorter text, or a lighter model variant. Source |
E6716 - Timeout Starting Prediction | The prediction did not start in time, often because of high load or slow model start-up. | Heavy load | Wait a moment and retry, or try off-peak. Source |
E8765 - Model Health Check Failed | The model's container failed its health check before the prediction started. | An outage | Try another model version or retry in a few minutes. Source |
Is it just you?
- Run a different, popular model: if it works, the problem is the specific model or its version, not Replicate.
- Cold starts can be slow; a prediction stuck in "starting" for a while is not always an outage.
- Check the error code: E1001 and input errors come from your input, while E1000, E6716 and E8765 point to the platform or model.
- Check replicate.com's status page and the reports on this page.
Alternatives to Replicate
Questions
Replicate returns four-digit codes starting with E. E1001 means the model ran out of memory, E6716 means the prediction timed out while starting, and E1000 is an unknown error; the error codes page lists them all.
Models that have not run recently need to boot, which can take a while. If many models are stuck at once, check Replicate's status page and the live status above.
Sources: replicate.com, replicate.com
Get alerted when Replicate breaks your app.
If your product depends on Replicate, your users feel its outages too. Point a Site Qwality monitor at the Replicate endpoint you use and at your own site, and we alert you on Slack, email, SMS or a webhook as soon as either one stops answering.