I’m using gemini 3 flash preview, thinking=low to solve my problems. But in recent day, gemini didn’t use thinking (return thinking_token=none), this made my system’s accuracy experienced a significant decrease. How can i make to solve this issue.
Hello @hung_hoang_dinh ,
To troubleshoot the issue, could you share the exact prompt and code snippet you are using, the SDK version, and whether this happens on simple inputs or during multi-turn conversations?