Replicate Uptime
Replicate API for running open-source models in the cloud.
Daily uptime
2026-04-29 to 2026-05-05Embed this uptime badge
Frequently Asked Questions
How is Replicate uptime calculated?
Every 2 minutes a Cloudflare Worker captures the status of Replicate from its public status feed. Each 24-hour window adds about 720 status samples per provider, 5040 over a 7-day rolling window. Uptime % is (operational_samples + 0.5 * degraded_samples) / decisive_samples * 100, where decisive_samples excludes unknown polls so brief gaps in our capture do not penalize the provider.
Why does degraded count as half?
Degraded service is not the same as unavailable. Most degraded periods (elevated latency, rate-limit pressure, partial-region issues) still let some traffic succeed. Counting degraded as half operational gives a fair single-number summary instead of treating all non-perfect time as equally bad.
Is the chart real-time?
It refreshes every 5 minutes on the page and every 2 minutes on the worker. The data underlying each bar is captured from Replicate's public status feed at that 2-minute cadence. Today's bar updates throughout the day as new poll cycles run.
What if today's bar shows no data?
If the chart shows a gray placeholder for any day, that means counter capture had no decisive samples for that day. New providers (or providers with status feed parser issues that have since been fixed) accumulate data from their first successful poll forward. Past days with no data stay empty by design.
Can I embed this uptime % in my README?
Yes. Drop this markdown anywhere: . The badge auto-updates as new poll data accumulates. See /badges for every provider's badge.
Where do I get historical data beyond 7 days?
The premium /api/premium/status/leaderboard endpoint extends to a 90-day rolling window and adds incident_count + mttr_minutes (mean time to recover) per provider for SRE/ops/procurement teams. See /developers/agent-payments for the paid tier (1 credit per call).