Gemini API stopped randomly in agent mode

Bug Reports
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

When using the agent mode with Gemini 2.5 pro (with my own API key) the agent mode stops mid thinking or does not think itself. This model becomes unusable with this as it has been occurring more recently. Any fix regarding this.

Steps to Reproduce

add you api key, then use the 2.5 pro agent based on your api key, after some usage it stops mid thinking or stop thinking itself. sometimes the thinking gets through and it stop mid file updates.

Expected Behavior

The model should think properly and generate response and update the files

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

Version: 1.4.6
VSCode Version: 1.99.3
Commit: 8bdbd55db6885f3d79f73cdacca5fcedef93d360
Date: 2025-08-15T02:08:45.149Z
Electron: 34.5.8
Chromium: 132.0.6834.210
Node.js: 20.19.1
V8: 13.2.152.41-electron.0
OS: Darwin arm64 24.6.0

Does this stop you from using Cursor

Yes - Cursor is unusable

2

Hey. Is everything okay with your internet? Maybe you’re experiencing network interruptions? Have you noticed this behavior when not using your API key?

3

internet is fine , as other models does not have the same issue, sonnet, gpt. I will test with it without api key and provide an update.

1 Like
4

@deanrie it is also breaking when using without the API key, it stops mid thinking or implementation phase, may be an issue at the google’s end. Read some reddit post’s of gemini breaking mid thinking or streaming.

added snapshot of it stopping mid thinking (without api key)

Screenshot 2025-08-20 at 22.36.30
Screenshot 2025-08-20 at 22.36.30498×384 27.3 KB

5

Yes, this issue is on Google’s end, and the team is already working on it with them.

1 Like
6

do we have any ETA or some update regarding this? It sometime works without the API key but with API key it just outright crashes.

7

experiencing same issue using gemini 2.5 pro via api key. pretty much unusable at this point