I am currently trying two different setups and wanted to share my experience and hear yours.
First, I used Cursor $200 on its own. Before that, I had already gone through Cursor Pro and the $60 plan. Even on the $200 plan, I consistently hit the usage limit about three to five days before my subscription renewed.
Because of this, I switched to a second setup: Cursor Pro ($20) + Claude Max ($200). I now use Claude Code inside Cursor, mainly because I have become very accustomed to this workflow over the past two years. Since switching to this setup, I have not hit a usage limit even once.
That said, there are still some things I miss from the Cursor-only experience. For example, I miss being able to easily select a div element in the browser preview and edit it directly with Composer. Sometimes I still do this out of habit and then have to explain the changes to Claude Code afterward for progress tracking. I also find the ability to use Claude Sonnet with a 1-million-token context window inside Cursor extremely powerful. In Claude Code, I am limited to the base 176k / 200k context size.
Right now, I am undecided between these two setups:
$200 Cursor vs. $220 Cursor Pro + Claude Code.
What has your experience been, and which setup do you prefer?
I don’t use Claude; I’m all-in on Cursor. The main reason I stay is the model variety and the fact that I almost never hit my usage limits. I’m not a fan of Opus, it’s usually overkill for daily work. I prefer using faster, more efficient models for my daily tasks.
My Cursor Setup:
Planning: GPT 5.2, Gemini Flash 3, or Sonnet 4.5.
Execution: Grok Code Fast, Composer, or Gemini Flash 3.
Complex Tasks: I use GPT 5.2 High for the heavy lifting.
(I only use Codex if I’m deploying to a VPS.) Most people over-rely on Opus, but as a regular engineer, I’ve found that Sonnet 4.5 or other “cheap” models are more than enough.
Your experience with Claude Max + Cursor Pro ($220 total) vs Cursor Ultra ($200) is an interesting case. The Cursor Ultra benefits you mentioned (1M context window with Sonnet, browser preview integration) are genuinely meaningful for your workflow.
We really appreciate discussions like this because they help the team improve Cursor’s workflow.
That is quite interesting. I have to admit that I use Claude Sonnet for daily tasks and Opus for planning and more complex tasks. If I do something like a swift cleanup of the root directory, consolidating documentation, etc., I use Composer because it is razor-fast. But it could be a good experiment to switch between different models more often, since Sonnet (1 million tokens) and Opus are quite expensive.
I’ve hit my limit with the $60 plan in like 3 days. I’ve now upgraded to Ultra, because Cursor let’s me build my project so much faster. I was critical about the price changes and even ended my subscription but came back eventually. I just accepted I can’t do this as fast without Cursor. Heck, i don’t even have to code at all anymore if i’m completely honest with you. Since i left 6 months ago Cursor and the LLM’s have improved so much that natural language is enough. It’s not as frustrating as it used to be. Opus 4.5 just blows my mind. I was used to errors after each change by the models but now it rarely happens anymore. So i wish to take back my earlier critique that the price change is unjustified. It’s absolutely worth it, even if it ain’t cheap.
AntiGravity is way out in front. It’s way faster making complex edits to multiple files. The way it uses the implementation plan, task list, and walkthrough artifacts to create cross thread memory is state of the art and the ability to edit them directly is sweet. It’s way more reliable with the terminal and can edit numerous file at the same time. It just works and works well. However, its MCP server support is buggy. It works with some and not with others. It is way, way cheaper, especially with the current sale. I work ALL I want to now with only Opus 4.5 thinking.
with anti gravity - 1 you dont know how many tokens you have - yes it kind of ok. but cursor has better ux (for me)
2 the good thing is your credits refresh every 5 hours if you pay $18 p/m.
3 i dont think they have open ai models
Interestingly, a friend of mine told me about it just yesterday, so I’m trying it out now. I started with Windsurf, then switched to Cursor, and later tried Kiro. I really liked the idea behind Kiro, but it seemed like a doomed project from the start due to the lack of commitment behind it. Now I’m trying Antigravity, and I’m extremely curious to see how it will perform.
At the moment, there is no real alternative to Cursor, but that could change quickly—especially since I’m quite unhappy with the direction Cursor is heading in terms of limitations and token usage. There was also a period of about a week when Cursor wasn’t working at all. Despite many users pointing this out and asking for a refund, nothing was offered. To be honest, that experience has shaken my trust in the company.
I believe we’re all early adopters, and if the company treats its early users this way, it doesn’t seem like customer satisfaction is a high priority. But that’s just my two cents.
i feel like mcps are less of a option with antigravity.
on the pricing side, they dont tell you how many credits you are using. they just give you a unkonwn amount that restes every 5 hours
not 100% sure how that works but it def cheaper. it is part of the google ai subscription.
also you have way less modles to choose from. (only gemini, claude and gpt-oss) so not necessarily the best
It’s way cheaper. Just using Cursor to test the MCP servers I build was costing a fortune. Now, I can work all I want with Opus 4.5 Thinking and I never get cut off. It’s on sale so the price is going up but that new price is still far, far cheaper than Cursor. I don’t care about token usage stats when it is so much cheaper. Greater visibility + greater expense doesn’t beat less visibility and less expense. I’d ignore the info anyway since usage is essentially unlimited at the moment. Cursor has more settings but are settings the same thing as features? Many of Antigravity’s features are nearly invisible and don’t require you to adjust settings but they work. But, not all MCP servers will work in Antigravity yet. I love Cursor but it’s too expensive and the bugs don’t help. It’s also much slower and less reliable for editing multiple files.
The usage limits on Claude in Cursor have been frustrating. I had one bad day with Opus–It just couldn’t solve any real problems and that one day blew all my tokens for the month. I really don’t want to switch and I also can’t afford to pay hundreds of dollars per month for Claude, so I found somewhat of a workaround:
Claude is good for one-shotting a problem. If you treat Composer the same, it doesn’t work as well. But if you micromanage Composer and make it explain itself before you let it act, it actually performs much better than it does as a one-shot model.
That is true. MCP is not where Cursor is yet. But I think this will change quite quickly. So far, I haven’t reached the limit at all. And I use it daily, just like I use Cursor. The fewer models don’t bother me. On the contrary, they offer Gemini Pro 3 High/Low, Gemini Flash, Opus 4.5, Sonnet 4.5, and GPT OSS 120B.
And in the end: I pay €102 instead of €180 – to me, that’s a strong argument.
But since Anthropic just released Claude Cowork, I can see myself switching again in a few weeks.
Hey Alex, yes, I understand. I used Claude Code too. I really love the quality of the output. The problem is, I am so used to the interface of an IDE that it is hard for me to switch completely. I always ran Claude Code inside of Cursor, but then I had $200 and $20 for Cursor Pro.