The temperature setting in Google AI Studio says “For Gemini 3, best results at default 1.0. Lower values my impact reasoning”.
I have realized that this is the biggest lie every told, especially for those that use gemini 3.1 pro preview for coding tasks.
Maybe 1.0 temperature may be better for people that do ‘creative writing’ but if you do coding and set it at 1.0, gemini 3.1 pro will take the liberty of changing an endless amount of stuff that you didn’t ask it to change, in most cases for the worst, creating bugs, removing logic, removing logs (for some reason this model really has a kink for removing logs from different parts of the apps… I guess it considers logs useless?), it will occasionally change position of widgets, or even change how the widgets are built, all for a prompt asking for a surgical small change.
Using the model with temperature set at 0.3 returns far better results and diminishes the chances that the model takes initiative and silently removes stuff from your code, breaks logic and so on.
While this is not a bug, it is at the very least quite a misleading statement to put as tooltip for the temperature setting.