Elastic Status · History · Incident #6374

RESOLVED

Increased error rates for microsoft-multilingual-e5-large

Major · Started Aug 13, 2026 · 8:32 AM

  • Duration

    4d 3h 8m

  • Severity

    Major

  • Detection lead

  • User reports

Summary

Increased error rates for microsoft-multilingual-e5-large

The microsoft-multilingual-e5-large model has been retired from the Elastic Inference Service. Requests to this model now return an error. Customers who wish to use this model with Elastic cloud must create a custom inference endpoint using their own API key. See our inference documentation for further details: https://www.elastic.co/docs/api/doc/elasticsearch/group/endpoint-inference


  • Started

    Aug 13, 2026 · 8:32 AM

  • Resolved

    Aug 17, 2026 · 11:40 AM

  • Duration

    4d 3h 8m

  • Severity

    Major

Event timeline

How this incident unfolded

  • Investigating

    Aug 13 · 8:32 AM Elastic

    We are seeing elevated error rates from the upstream provider for the microsoft-multilingual-e5-large embedding model. Search embedding requests to this model may fail. We are investigating

  • Investigating

    Aug 13 · 2:45 PM Elastic

    We're seeing some improvement in availability but are continuing to monitor the situation. We will follow up with more information in due course.

  • Monitoring

    Aug 13 · 6:26 PM Elastic

    Unfortunately, we can no longer support the E5 Multilingual Large model on the Elastic Inference Service due to continued upstream provider issues. Any customers requiring this model should set up custom inference endpoints using their own API keys.

  • Resolved

    Aug 17 · 11:40 AM Elastic

    The microsoft-multilingual-e5-large model has been retired from the Elastic Inference Service. Requests to this model now return an error. Customers who wish to use this model with Elastic cloud must create a custom inference endpoint using their own API key. See our inference documentation for further details: https://www.elastic.co/docs/api/doc/elasticsearch/group/endpoint-inference

Get alerted before the next Elastic outage.

Pulsetic catches degradations minutes before vendors acknowledge them.