This incident has been resolved.

Switching models has stabilized Neo inference calls

We have switched model versions and are continuing to monitor

We are currently experiencing a high rate of upstream errors for Neo inference calls.