Thinking models are great for accuracy and it seems that newer models all seem to have it onboard by default. It would be great if there was a filter option to disable the thinking output to the screen etc as part of the response.
Hi,
As Gemini 2.5 models are performing better than Gemini 2.0. models, I like to use Gemini 2.5 within voice assistant.
However every response when using Gemini 2.5. models starts with my voice assistant ( ATOM Echo) spitting out the thinking process before actually answering, which is bit tedious.
Is it possible to refrain or by config include/exclude the thinking process. See eg example of thinking process below:
The user is asking for tips on visiting Amsterdam. I need to generate search queries in Dutch to find relevant information about visiting Amsterdam.
1. Generate search queries:
** Wat zijn tips voor een bezoek aan Amsterdam?*
** bezienswaardigheden Amsterdam tips*
** Amsterdam toeristische tips*
** Wat te doen in Amsterdam?*
Hope this can be considered. For now I simply use Gemini 2.0. as that doesnt include thinking process in the response.
Agree. Seeing the model’s thinking is frustrating for implementations with devices like the Voice Preview Edition where responses are spoken via TTS.
Update: I added the following text to the end of my prompt and it seemed to remove the chain-of-thought reasoning (thinking) text passed to TTS in my voice assistant pipeline. Unclear if this stops the model from thinking entirely or just hides the chain-of-thought reasoning from the output.
/No Thinking
1 Like