Resolved incident

Degradation with Gemini 3.8 Flash

GitHub · Major · Started

Duration
10h 28m
Severity
Major
Status
Resolved

Summary

Major · Resolved

On September 16, 2026, between 04:40 and 11:45 UTC, the Gemini 3.8 Flash model in GitHub Copilot experienced degraded availability. Requests to this model failed at an average rate of 6.4%, and the impact was highest during peak traffic hours. Other Copilot models were not affected. Users could continue to work with a different model or with 'Auto'.<br /><br />The degradation was caused due to a capacity issue with an upstream model provider. Failure rates returned to normal as the daily traffic peak passed. We monitored the model until it was healthy and resolved the incident at 17:48 UTC. We are working to make our systems resilient to cover peak demand for all the Copilot models.

View official incident ↗

Event timeline

How this incident unfolded, from the official status page.

  1. Investigating

    GitHub

    We are investigating reports of degraded availability for Copilot AI Model Providers

  2. Investigating

    GitHub

    Copilot AI Model Providers is experiencing degraded performance. We are continuing to investigate.

  3. Investigating

    GitHub

    Copilot AI Model Providers is experiencing degraded performance. We are continuing to investigate.

  4. Monitoring

    GitHub

    The degradation affecting Copilot AI Model Providers has been mitigated. We are monitoring to ensure stability.

  5. Resolved

    GitHub

    On September 16, 2026, between 04:40 and 11:45 UTC, the Gemini 3.8 Flash model in GitHub Copilot experienced degraded availability. Requests to this model failed at an average rate of 6.4%, and the impact was highest during peak traffic hours. Other Copilot models were not affected. Users could continue to work with a different model or with 'Auto'.<br /><br />The degradation was caused due to a capacity issue with an upstream model provider. Failure rates returned to normal as the daily traffic peak passed. We monitored the model until it was healthy and resolved the incident at 17:48 UTC. We are working to make our systems resilient to cover peak demand for all the Copilot models.