I've become extremely nervous about these sudden declines in quality. Thankfully I don't have a production product using AI (yet), but in my own development exp

I've become extremely nervous about these sudden declines in quality. Thankfully I don't have a production product using AI (yet), but in my own development experience - the model becoming dramatically dumber suddenly is very difficult to work around. At this point, I'd be surprised if the different vendors on openrouter weren't abusing their trust by silently dropping context/changing quantization levels/reducing experts - or other mischievous means of delivering the same model at lower compute.