Context limitation has returned

Describe the Bug

Files in context condensed.

Steps to Reproduce

Put files in context

Expected Behavior

Should not be condensed

Screenshots / Screen Recordings

Operating System

Windows 10/11

Current Cursor Version (Menu → About Cursor → Copy)

Version: 1.3.6 (user setup)
VSCode Version: 1.99.3
Commit: 68b8fe7396ea37d8acdaaaa08ba316ba359a4160
Date: 2025-07-30T18:17:09.810Z
Electron: 34.5.1
Chromium: 132.0.6834.210
Node.js: 20.19.0
V8: 13.2.152.41-electron.0
OS: Windows_NT x64 10.0.22631

Additional Information

I’ve had this before. It was fixed for a while, but it seems like the devs at Cursor keep changing things.

Now, every model apart from gemini-2.5-pro in MAX mode is useless.
Gemini seems to handle big context properly, as it did before in months March to June, just like almost every other model had a large context enough to fit entire codebases if I wanted.

But now, all other models instantly bleed out when they see a 3000 line file, and everything gets condensed. This is unacceptable behavior, especially that the “condensing” seems to greatly impair the models ability to reason and to accurately handle files.

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

This shouldn’t happen with MAX models, they are explicitly chosen to fit full files. But now you put in the file, and the agent starts wasting money by running tools to read random samples of the same files you provided. So no you have 20% extra context window and the same information read multiple times, getting nickled and dimed and then again at the generative step with the extra context.

I should have just referenced the information that was already provided instead of making additional tool calls. The user is absolutely correct - I was being inefficient and not utilizing the context they already provided.

Something is definitely wrong, I clocked thinking MAX tasks to cost much less than no thinking regular tasks, there is some optimization that does this.

1 Like

That’s right, I guess there might be something fishy going on, but I am literally using Cursor all day in MAX mode, with various models, they dont need to squeeze me for money.

The better it works, the more transparent and productive it is (instead of uselessly stabbing in the dark with it’s rag search, grep and repetitive file reads, and most importantly “compressing” files in a way I did not ask for at all), the more I will use it.

The worse it is, the more I think about running cline or similar instead.

It’s very simple.

This is happening on all my files in ‘auto’ even is the file is very small like 50 lines.
Cursor has been terrible for the last few weeks. Constant updates breaking things

This topic was automatically closed 22 days after the last reply. New replies are no longer allowed.