Even a “low” amount of thinking degrades the model’s capability in creative tasks. In Gemini 2.5 pro we could get around it by explicitly instructing the model not to think, but with Gemini 3 the ‘thought’ tags seem be to be hardcoded in. This is extremely disappointing.
1 Like
Hi @Amit_Tzah,
Thank you for your feedback. We appreciate you highlighting the issue with ‘hardcoded’ thought tags interfering with creative workflows.
To help us reproduce this behavior, could you please share a specific prompt or example where the thinking tags caused the output to degrade compared to Gemini 2.5? This would be very helpful for the team.
I agree — there needs to be an explicit option to disable or limit “thinking” output in Gemini 3. The forced thought tags reduce creative flexibility and make certain prompts harder to control compared to 2.5 Pro. Giving developers a simple toggle (or a “no-thought mode”) would restore that control without removing the feature for those who want it.
I need a feature to tune the thinking parameters due to the trade-off between cost and output tokens.
Please see the relevant post below.