Sometimes it happens that in ai studio, even with thinking set to “HIGH” the model doesn’t think when answering, and instead just immediately outputs code. This should not be happening as far as I understand it.
This happened both with temperature 1.0 and also with lower values like 0.3
No other tools were selected, all other options were the default ones.
When this happens, usually the code quality is lacking or it may lack some context from previous replies. It definately feels like unintended behavior.
This is hapening almost all the time with all models. This way they can save money but models are nerfed hard almost unusable. And that is going on since december btw
Yes, it appears they added some sort of “if the query is not too complex → skip thinking”, but this often fails to assess the actual complexity of the query and also it is not fair to users that explicitly select THINKING and expect a certain result.
I understand that this might be a decent optimization for cases where people use big models like 3.1 pro preview with thinking HIGH to do basic tasks like translating a text or creating writing, but if the task is coding related and thinking HIGH is set, it should never skip thinking.
nothing to do with the outage (which I do not see, as it is working as usual today).
The issue happened some days ago and then again yesterday. And it also happened different times in the past, as the user @hary_canto mentioned, since end of december/ january.