Hello,
I’ve noticed a serious issue that should be addressed as a bug. Although I haven’t changed the model type I’m using, it clearly switches without my request and without any notification.
This is especially noticeable when using thinking models like Sonnet 3.7. with thinking enable. The model suddenly stops “thinking” through answers—it no longer includes any step-by-step reasoning or explanation, which was a key part of its behavior before.
This kind of silent change is problematic. Not only does it affect the quality and reliability of the experience, but it also breaks user trust. If the model being served is different from what’s selected, it should at the very least be communicated clearly.
Please investigate this as a potential bug and consider implementing safeguards or notifications for model switching.
Thank you
Avi