Vendor record
Replicate
Checked
Front door check counts and Replicate’s own incident posts. Use them to investigate a dependency; neither confirms what happened in your app.
Front door checks
We send a request to a documented Replicate endpoint and count the checks that got the answer it is documented to give.
Front door reachable in 2,879 of 2,879 checks with a clear answer, last 30 days, checked every 15 minutes from a single CheckUpstream location.
- Endpoint checked
- api.replicate.com
- Scope
- apiNo vendor region
- First check in this window
- Latest clear answer
Checks 15 minutes apart cannot see disruptions shorter than that. A check that reached the front door does not show that every Replicate product, region or your app worked.
How we count front door checksReplicate says
Incident posts from Replicate’s status page, with its titles, impact levels and times. They say what Replicate chose to post, not what reached your app.
A closed record shows when our record closed. Some older records closed when a post left the status page, without a resolution post from the vendor.
19
incidents Replicate posted, last 1 year
Median recorded duration: 4h 19m. From 16 incidents with vendor-supplied starts and recorded closure times.
Most recent records · all dates
Replicate Flux Hotswap Models Stuck
MinorReplicate’s postStarted, per Replicate
Closed in our record(19m)
Intermittent 500 errors for backend services on Replicate
MinorReplicate’s postStarted, per Replicate
Closed in our record(1h 59m)
Replicate Pruna issue
MinorReplicate’s postStarted, per Replicate
Closed in our record(20h 45m)
Some Replicate models are failing to start
MinorReplicate’s postStarted, per Replicate
Closed in our record(15h 44m)
A100 hardware partial outage
Not specifiedFirst seen by CheckUpstream
Closed in our record
14 older records
Delayed scaling due to node failure
Not specifiedFirst seen by CheckUpstream
Closed in our record
Significant degradation
Not specifiedFirst seen by CheckUpstream
Closed in our record
API degraded for A100s
MajorStarted, per Replicate
Closed in our record(10m)
Degraded scale-out due to failed setups pulling from huggingface
Not specifiedStarted, per Replicate
Closed in our record(2h 9m)
Hitting GPU Capacity for H100s creating large queue times for some models
Not specifiedStarted, per Replicate
Closed in our record(1d 2h)
HuggingFace download issues
Not specifiedStarted, per Replicate
Closed in our record(7h 24m)
H100 GPU shortage resulting in high queue times
Not specifiedStarted, per Replicate
Closed in our record(1d 1h)
High contention on H100 hardware
Not specifiedStarted, per Replicate
Closed in our record(8h 15m)
Limited H100 capacity
Not specifiedStarted, per Replicate
Closed in our record(5h 29m)
We're seeing long setup times and high contention for models on some L40S and H200 clusters.
Not specifiedStarted, per Replicate
Closed in our record(1h 13m)
Degraded performance on flux-2-klein-4b
Not specifiedStarted, per Replicate
Closed in our record(1h 49m)
Prediction and Training status updates delayed
Not specifiedStarted, per Replicate
Closed in our record(1h 50m)
Constrained H100 capacity
Not specifiedStarted, per Replicate
Closed in our record(7h)
Constrained capacity for H100 hardware
Not specifiedStarted, per Replicate
Closed in our record(4h 19m)
Your code, your dependencies
See which of your code paths call Replicate
Connect GitHub or GitLab to find where your public or private repository uses Replicate. Review each match and your alert settings.
Free to start. No credit card.
Use this record on your site
Add Replicate status to your page.
Choose the format your page supports, then copy the code. Badges show the recorded vendor status; they do not establish impact on your app.
Frequently Asked Questions
Is Replicate currently down?
Replicate's latest recorded status is Operational. Check the timestamp and linked source at the top of this page. A vendor status report does not establish whether your application was affected.
When did Replicate last have an outage?
The most recent closed incident record began 17 days ago and spans 19 minutes. The vendor reported degraded performance. Closure may reflect a disappeared post or an expired tracking period; it does not confirm vendor or application recovery. Check the dated history and source links.
Does CheckUpstream publish an uptime percentage for Replicate?
No. CheckUpstream checks Replicate’s front door every 15 minutes and publishes how many checks reached it over the last 30 days, with the endpoint checked. Checks 15 minutes apart cannot see disruptions shorter than that. A percentage would claim more than the checks can show.
How do I monitor Replicate in production?
Create your account, connect GitHub or GitLab and choose a public or private repository. Review the vendor matches and your alert settings. Available notification channels and history depend on your plan.
How does CheckUpstream track Replicate?
This record separates vendor-reported incidents from independent checks. Sources and coverage vary by vendor: review the linked source, check time and measurement scope. Neither kind of evidence establishes SLA credit eligibility.