Is Replicate Down?
Check live Replicate status — predictions API, model warm-up, and GPU availability. See recent incidents and set up free email alerts.
Replicate — live status
Updated every 5 minutes. Full incident history at prismix.dev/service/replicate.
Quick check: is Replicate down right now?
- Prismix: prismix.dev/service/replicate — live status + 30-day uptime + incidents.
- Replicate status page:
status.replicate.com— official Replicate incidents. - API call:
curl https://prismix.dev/api/v1/statuses | jq '.services[] | select(.id=="replicate")'
Set up free email alerts for Replicate
- 1
Sign in
Go to prismix.dev/sign-in — email OTP or GitHub sign-in.
- 2
Star Replicate
On prismix.dev/service/replicate, click the ☆ star icon.
- 3
Alerts are live
You'll get an email within minutes of any status change.
Monitor Replicate programmatically
curl
{`curl https://prismix.dev/api/v1/statuses | jq '.services[] | select(.id=="replicate")'`} Python
{`import httpx, sys
resp = httpx.get("https://prismix.dev/api/v1/statuses", timeout=5)
svc = next((s for s in resp.json()["services"] if s["id"] == "replicate"), {})
if svc.get("indicator", "unknown") != "none":
print(f"⚠️ Replicate is {svc.get('indicator')}: {svc.get('description')}")
sys.exit(1) # fail CI or skip generation`} Common causes of "Replicate not working"
If Prismix shows Replicate as "Operational" but predictions are failing or slow:
- Model is in cold start (slow) — Replicate unloads infrequently-used models from GPU memory. A prediction that triggers a cold start can take 30–120 seconds before producing output. This isn't an outage; poll the prediction with exponential backoff using
GET /v1/predictions/{id}. - Rate limit exceeded (429) — Free tier has 50 predictions/minute. Upgrade to Starter or Pro for higher limits.
- API token missing or invalid — All prediction endpoints require
Authorization: Token <REPLICATE_API_TOKEN>. Generate at replicate.com/account/api-tokens. - Model version not found (422) — Replicate model versions are immutable and may be deprecated. Always pin to an explicit version hash, not
latest. - Prediction stuck in "starting" state — If a prediction stays in
startingfor 5+ minutes, it may be in a GPU queue. This correlates with Replicate's degraded/high-demand events. - Webhook delivery failures — Replicate delivers async prediction results via webhooks. If your endpoint is down, predictions still succeed but you won't receive the callback. Store the prediction ID to poll manually.
Get an email the next time Replicate goes down
Outage alerts for Replicate, straight to your inbox. No account needed, unsubscribe in every email.
Watching more than one service? A free account covers 5 services + a daily digest — and Pro is currently free.
Monitor other AI model hosting providers?
Full status dashboard: prismix.dev/status