Minor model quality issues expected for a few more days

I am currently having issues with one of my service providers. I believe I've already resolved model quality issues for all requests except Unified mode.

For unified, I'm using a backup model which performs about as well as the preferred model when it works, but doesn't follow instructions quite as accurately and has the tendency to randomly produce wrongly formatted text about 1% of the time. Many of you may have already noticed this. Sorry for the inconvenience. I'm hoping to be migrated to the preferred model within 1 week. For now, retrying the failed query should help as a workaround.

Please feel free to continue to report issues involving extremely long response times, non-Unified requests, or persistent/game-breaking issues with the Unified model.

Thanks, Max

Update 05/19/26: As a further mitigation I've managed to reduce the number of errors by ~95% using server-side retry logic. Errors should be pretty rare now