We've confirmed that the issue has been resolved with our upstream provider and error rates have remained stable at zero.

Starting Thursday, June 4th at approximately 19:12 UTC, we experienced an issue with OpenAI image and video models (GPT Image and Sora) due to a provider-side issue. Requests that used any of these models would fail. This also impacted Flows and Templates. This was fully resolved at 19:52 UTC on the same day.

The incident has been resolved. The root cause of the problem was a GCP incident affecting I/O on certain VM types in one of the zones we are using. As a result, some nodes in our clusters were affected.

The incident was resolved.

The issue has been resolved, and exports should now be working as expected.

Our billing provider has corrected the problem, and all subscription updates are processing normally and without delay.

The issue affecting Text-to-Speech (TTS) generation, which was causing intermittent 500 errors, has been resolved. The service is now operating normally.

The issue has been resolved and error rates are back to normal.

The issue has been resolved. We will publish a public write-up on this status page once the investigation is completed.

We have implemented a fix for the issues affecting EU residency STT requests that use the source_url parameter or the deprecated cloud_storage_url parameter and confirmed full issue resolution.

We've tracked error rates and latency as they have returned to baseline, and confirm full resolution of the ASR issues affecting the EU Residency region.

We have released an update to resolve this issue.

We have confirmed full issue resolution.

The issue has been resolved, and the system is now showing the correct message.

We have taken measures to restore full SIP functionality and tracked error rates as they have returned to baseline.

The issue has been fixed; this affected only newly created accounts today.

The issue has been resolved, and all the voices should be visible now. This only affected workspaces with 50,000 or more voices.

The root cause was an incompatibility in how ElevenLabs handled mid-dialog re-INVITEs. When SIP providers sent re-INVITEs, ElevenLabs responded with 488, which some providers treat as fatal, issue a BYE, and terminate the call. Per RFC 3261, a 488 should not necessarily result in session termination, but provider behaviour varied.

The issue has been fully resolved.

We were able to identify the root cause of the increase in errors. We have implemented a fix to mitigate any possible occurrences moving forward.

EU Residency:\n\nStarting Saturday, March 14th at approximately 17:00 UTC, some calls were affected by connection failures, or connected with silent audio. This was fully resolved at 08:00 UTC on March 16th.\n\nCross-Region:\n\nStarting Sunday, March 15th at approximately 01:00 UTC, some conversations experienced a 1 second delay to initial audio. In some cases, this caused disconnections due to how some SIP trunks handled the audio delay. This issue was fully resolved at 16:00 UTC on March 16th.

Our team has resolved the issue, services are back to normal.

We have confirmed with affected clients and in our analytics that SIP trunking functionality is now fully restored.