I’ve recently made the difficult decision to switch from Cursor to Windsurf, despite having already paid for an annual Cursor subscription. I want to share constructive feedback in the hope it helps the team understand why users like myself are moving to competitor products. My decision comes down to two key issues:
Severe Performance Issues (Freezing/Locking Up)
Cursor has always struggled with performance as chat/context windows grow, but since v1.3, when the chat layout was changed and ‘current tab’ context was introduced, the freezing has become constant, even with minimal context in the chat window. The app now locks up after every few words I type, making it virtually unusable. My current workaround is to draft responses in Notepad and paste them in during the brief moments the app is responsive.
(Yes, I’ve tried clearing the workspaceStorage folder, it helps for a few minutes, then the freezing returns after a couple of back and forth interactions with the chat).
Buggy Releases & Lack of Stability
Cursor feels like a beta product. It seems releases are pushed with minimal QA or testing, and I often feel like I’m part of an involuntary beta program. For a tool used in professional workflows, stability must take priority. The team needs to adopt a stable release branch and isolate experimental features to optional alpha/beta builds. No new features should be shipped until the app is stable.
I started with Cursor because they were first to market, but after months of frustration, I’ve now moved on, and for what I need, Windsurf seems near identical, and it feels more stable and doesn’t constantly freeze.
I hope the team will take this feedback seriously and refocus on performance and stability.
We are also seriously thinking about doing the same. We also have the yearly Cursor Pro, but endless daily issues plague it. Out of curiosity, as you’re using Windsurf as we speak, is it possible to use it with our APIs from open-source LLMs like GLM 4.5, Qwen3, etc?
It’s Windows 10 and, granted, the PC is pretty old, but Windsurf seems to work fine… Sorry, I don’t mean to bag on you guys, I’d be more than happy if you fixed and improved Cursor, but I literally can’t use it - the majority of the time (I’m not exaggerating) it’s frozen / unresponsive. Obviously I can’t work like that.
In fact there is a oneliner reason for leaving Cursor: their silent change to “MAX MODE” now compresses the context, so you can’t use Gemini 2.5 Pro 1 Million tokens context in Manual Edit mode (even if you are ok with the requests cost of 1M tokens) … this is the most ridiculous non-sense absolutely non-negotiable (and their removal of Manual Edit mode feature); I think someone is trying sabotage Cursor, they can’t be that du*mb in such a short period of time?
(and I think moving to Windsurf won’t solve the problem, as they have a non-configurable max 64k tokens context window per request?)
As much as i don’t like the new pricing model, i’ve tested alternatives like kiro, trae, and windsurf and nothing comes close to cursor when it comes to trusting it with your codebase in terms of making the correct edits. windsurf and trae operate as if they know nothing about your code. I have active plans with both, you could call me a power user, but the only one that consistently gets things right is cursor. It’s a bummer, and honestly quite sad, that nothing else is close right now. you could go with Claude Code, but it’s pretty expensive some might even say it costs more than gold.
Not our experience here. Windsurf has issues but Kiro is absolutely fantastic; we use it a lot. The only downside is that it does not work (yet) with API.
I completely agree! Zero stability, it helped me to throw the folder with all chats and similar stuff into the RAM. I have already canceled my subscription, and the only thing that kept me here was the incredible suggestions, but the competitors are already much better, so I can switch to them. No one is requesting 500 requests like before; everyone is requesting stability. Simply stability and acceptable performance. No, we do not need 1,000 new features; please conduct proper QA before releases.
I am also evaluating other products for the same reason. For me, I keep getting communication link errors that cause the AI to stop working, sometimes mid-change, leaving my code in an inconsistent state.
I do appreciate the excellent support by the Cursor team, and how well the Cursor team listens to its users and responds to feedback with new features, but losing days worth of work over stability issues is kind of a deal breaker. I am confident that the Cursor team will be able to resolve these issues, but I’m not sure that I can afford to wait for that to happen.
