Model Predictions Stuck at "Starting"
A record of the Replicate incident on February 20, 2026 (Japan time): start time, duration and severity, plus current status, what to do, and alternatives.
Other providers down at the same time
No other provider we monitor recorded an incident during this window. This appears to have been specific to Replicate.
Derived from our own archive of official status pages across 13 providers. Items lasting more than 24 hours and unresolved items are excluded, as they would misrepresent the moment.
How big was this outage?
This outage lasted 60 minutes. Across 12 major incidents recorded for Replicate, the median duration is 182 minutes, making this the 11th longest.
Shorter than a typical outage.
Only incidents recorded as major-or-worse and already resolved are counted.
Current status
Replicate is operating normally now. This incident has been resolved.
See current Replicate status →What you can do during an outage
Source: Replicate's official status page.
Other Replicate incidents
- A100 hardware partial outageAugust 27, 2026
- Delayed scaling due to node failureAugust 11, 2026
- Significant degradationAugust 5, 2026
- API degraded for A100sAugust 3, 2026
- Degraded scale-out due to failed setups pulling from huggingfaceAugust 1, 2026