I just tested using gemini Flash-Lite-Latest model and it’s working fine.. Please check if you are still facing this issue.
I just tested thinking_budget=0 on gemini-2.5-flash-preview-09-2025 and it continues to output thinking tokens, as discussed here.