How thinking budget and thinking works in gemini 3.1 flash image preview model

Gemini API
,
1

I am using gemini-3.1-flash-image-preview for generating images. In the doc it mention thinking is enabled by default, but if log usage metadata i see thinking token cost is zero. so if i add thinking_config, include_thoughts=True, thinking budget=2048, i get the thoughts back in my response. I want to know how does it work? By allowing thinking budget would my model think and generate better images? or the models thinks anyway. It just returns the thoughts when i add budget and include thoughts? Also in the docs it doesn’t mention how does thinking token cost.