Gemini 4 Argon Leaks Thoughts in Chat

Support Bug Reports
,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Gemini 4 Argon consistently leaks its internal thoughts into the chat. I’m not sure if this is a problem with the model or a problem with the Cursor IDE. Here is an example of this same issue happening to someone else, but in the Gemini web application - Reddit.

I know that technically this report would be better handled by someone at Google, but since I’m using the model through Cursor I’m not sure how I would report it to them. I assume y’all have contacts at Google you could work with?

Steps to Reproduce

Just use Gemini Argon on medium thinking level in the Cursor IDE, and give it some difficult tasks that require some thinking. After a while, you’ll notice that it leaks its thoughts. It likes to use the word “wait” when this happens.

Expected Behavior

I shouldn’t see those thoughts in the chat. They should be in the hidden “thinking” sections, or they shouldn’t appear at all.

Operating System

MacOS

Version Information

IDE 3.29.9

For AI issues: which model did you use?

Gemini 4 Argon

Does this stop you from using Cursor

No - Cursor works, but with this issue

5

Hi @George_Berdovskiy, this isn’t coming from anything in your setup. The thinking section only holds what the model sends back marked as thinking. As far as we can tell, Gemini 4 Argon sometimes writes its reasoning into the answer itself instead, so it shows up as normal chat text. Seeing the same thing in the Gemini app points the same way.

Switching Argon to the High thinking level is worth a try for now. It does more of its reasoning in the hidden thinking step there, though I can’t promise that removes it completely.

Could you share a screenshot of one leaked passage and the Request ID from that chat (the ⋯ menu at the top right of the chat → Copy Request ID)? We can’t see chat contents on our side, so one example would tell us a lot.