Major incident

UiPath: Gemini-2.5-Flash not working in Australia

Affected: IXP

Current status of UiPath: Partial outage — live UiPath status

Timeline

1. Oct 22, 2026, 12:59 UTC
   investigating
   We have identified an issue affecting IXP for customers hosted in the Australia region. Since 10:24 UTC, requests that use the Gemini 2.5 Flash model are failing in the region. Other models are not affected. Customer data continues to be processed only within Australia. We are working with Google to restore service
2. Oct 22, 2026, 14:19 UTC
   monitoring
   A fix has been deployed and the service is recovering. We are monitoring the system to ensure full recovery. Further updates will be shared soon
3. Oct 22, 2026, 14:33 UTC
   investigating
   Following initial recovery, the service has degraded again. We are investigating and working towards a resolution. We will provide further updates as they become available.
4. Oct 22, 2026, 15:42 UTC
   investigating
   We've contacted Google Support, and they've started investigating on their end. We'll keep monitoring and share updates as we have them.
5. Oct 22, 2026, 17:56 UTC
   monitoring
   We are observing improvements on Gemini 2.5 Flash traffic. We are monitoring a possible fix from Google.
6. Oct 22, 2026, 21:02 UTC
   resolved
   Issue is resolved from Google.
7. Oct 26, 2026, 03:20 UTC
   postmortem
   ## Customer impact Between October 22, 2026 at 10:24 am UTC and October 22, 2026 at 6:04 pm UTC, document processing requests that used Gemini 2.5 Flash in the Australia region failed with capacity errors. The impact lasted approximately 7 hours and 40 minutes. A subset of customers using document extraction workflows in that region were affected. During a temporary in-region fallback attempt, some extraction requests recovered briefly, but word-level attribution and generated page text were degraded until the fallback also began failing. ## Root cause The root cause was a provider-side regional network disruption during concurrent maintenance. This caused severe bandwidth throttling in the provider's model-serving environment, which prevented the backend systems from loading the model data needed to serve Gemini 2.5 Flash and Gemini 3.5 Flash requests reliably in the affected region. ## Detection The issue was detected through automated alerting that identified rate-limiting errors on Gemini 2.5 Flash in the Australia region. Our engineering team confirmed through telemetry that affected requests were failing consistently and began coordinating a response. ## Response At 1:40 pm UTC, we routed affected Gemini 2.5 Flash traffic to Gemini 3.5 Flash within the same region to preserve data residency. Initial telemetry showed no capacity errors after that change, but by 2:10 pm UTC, Gemini 3.5 Flash was also returning capacity errors, so the fallback was no longer viable. At 2:05 pm UTC, we adjusted related document processing paths so attribution and page-text generation would use the same in-region fallback; that adjustment completed at approximately 2:09 pm UTC. After the fallback also failed, we reverted the routing change, and by 4:28 pm UTC traffic was back on the expected Gemini 2.5 Flash path. At 5:38 pm UTC, telemetry showed improving Gemini 2.5 Flash success rates. From 6:04 pm UTC onward, no further capacity errors were observed, and subsequent requests succeeded. ## Follow up 1. Complete the reserved model capacity assessment for the affected region using the observed traffic patterns 2. Evaluate retry and backoff behavior in document processing clients to reduce retry amplification during provider-side capacity failures.

Updates as published by UiPath (original).