I’m completely non-technical. Here are some behaviors I’ve had in the PAST with Cursor:
- As conversations get longer the context gets muddied and becomes impossible and the model becomes significantly worse. Instructions are ignored and creating a new chat is the only way to continue.
Here is my experience after Sonnet4 and recent updates:
- The longer a conversation goes on, the more refined the model’s understanding of the codebase gets. It becomes BETTER. Creating a new chat often leads to that important “understanding” being lost. This is the complete opposite of what I experienced before. While code quality often isn’t perfect, Cursor’s agent-management of the context window has become stellar.
Just wanted to share since now I can effectively do the things I want to in Cursor. I was struggling for months before and now I’m actually able to implement features and see my codebase work how I want it to. Thank you!
4 Likes
Appreciate the love - it’s great to know our improvements and updates are genuinely improving the workflow of people on all levels of technical background! 
3 Likes
What was the model you used before? And how much of the “update” is actually better models available?
1 Like
I used Sonnet3.7 before and now I use Sonnet4.
It is absolutely due to the model - less hallucinations, etc.
But you’re missing the key factor and its that Cursor Agent provides the dynamic context window and tooling for the model. I have a Claude app plan as well and recently downgraded because Cursor’s context window and tooling management helps me get things done.
TLDR Model plays a large part but its Cursor that makes the difference for me!
2 Likes
technical or not 2jfs904judsw20600jik, from an AI dork on the other side, it is heavily noticed on this end too. clear step up - the Auto function particularly has gotten stronger.
Glad to hear that. This should be the experience for everyone. Hopefully, one day it will.