Probably been discussed before - but I’m on pro and I just went past $100 in overages. Using 4.5 Opus a lot to get through complex issues that Sonnet gets stuck on. I don’t mind paying for it - but curious if it makes sense to just keep paying overages or if the pro+ or ultra plans are a better bang for the buck (at least up to the point of maxing their limits). I can’t find a good idea of how much usage I get for my $20, 60, $200 vs. dollar for dollar on overages. I asked Claude and ChatGPT (both in research mode) and they both pretty much said ¯\(ツ)/¯
20usd plan = you get 20usd worth of usage
60usd plan = you get 70usd worth of usage
200usd plan = you get 400usd worth of usage
I personally switched to ultra, each month I use about 150-400usd worth of usage, some months I “pay more than I use”, some months I get “more than what I pay for”.
Also I am not sure if this is still real, but there used to be “extra allowance” beyond your subscription limits, e.g. on 60usd plan (70usd usage promised) you could get 100usd worth of usage. If anybody has any information on this, feel free to correct me.
1 Like
Ultra gives you $400+ for $200. If you spend or plan to spend more than $200 per month on Cursor, it’s worth it.
3 Likes
How does this compare to BYO anthropic key?
I’m over on my Pro plan in cursor, looking to upgrade to Ultra but seeing conflicting information regarding BYO anthropic key vs ultra.
1 Like
Claude API pricing is pay-as-you-go and quite expensive. I’d use $200 Cursor before that, and get the extra usage (I think I broke $500 in usage the last month I used it). Then I switched to Claude Max (with Claude Code) for $100, and it is pretty close to unlimited Opus usage in practice. I don’t even bother switching models anymore. I’ve been using it inside and outside of Cursor and dropped my Cursor sub to $20 mostly to use Grok Code to spin up debugging sessions and commit code, and the occasional use of GPT-5.2 High for extra debugging insight when Opus starts to struggle and needs another opinion.
I am hoping Cursor finds a way to allow Claude Code as a tighter integration (the Claude Code VSCode extension isn’t bad but it can’t use --chrome yet, and it is the best browser-agent-dev experience I’ve had so far.
2 Likes
Thanks for all the advice. $400 for $200 - that’s the info I was looking for. Worth the upgrade, I think - but then again, I’m paying for the $100 Claude subscription too and it never seems to run out. But there’s one problem: Claude Code is awesome until it starts compacting (or at least as of last month when I used it heavily still). Then it gets full on dementia and can’t even seem to remember the last few turns correctly. Cursor does such a good job of maintaining coherence when the conversation starts getting compacted that I mostly stopped using Claude Code for the past month.
In the end - paying $300 a month for both to do weeks of work in a day is still worth it no matter how I slice it.
1 Like
I am trying the same thing at the moment. I started working with Cursor, upgraded to Pro, then $60 Cursor, then $200 Cursor. Still, I hit the limit about three to five days before my subscription renewed. So I downgraded to Pro ($20) and upgraded to Claude Max ($200). I use Claude Code inside Cursor, since I have gotten really accustomed to it over the past two years. Since I switched, I have not once hit a usage limit.
Still, I am wondering what your experience is. I realize that I miss the ability to easily choose a div element in the browser preview and edit it with Composer (sometimes I still do that and then have to tell Claude Code what I did for progress tracking). Also, the ability to use Claude Sonnet with a 1-million-token context window is pretty powerful in Cursor. In Claude Code, I only have the base 176k / 200k context size.
So right now I am undecided about the question: $200 Cursor vs. $220 Claude Code and Cursor. What do you guys think?
I use claude --chrome with the chrome extension installed. It is an MCP to its own browser extension. I often let it do work and check results until done this way. Sometimes I will screen shot a dev-tools element, but honestly it just figures it out if I explain it and tell it to look in chrome. This is way easier to use than the Cursor browser since it is just connecting to a regular chrome instance.
@cursor maybe make an chrome extension so you can do the same? I still very much want to launch a debug session from Cursor and also have the models see the same instance. I don’t think anyone does this yet for me.
Regarding the /compact note above, it isn’t really different than Cursor’s /summarize. It shows you it is coming and it is almost always better to just compact/summarize when changing subjects slightly, or just open a new terminal tab or /clear to start totally fresh. I usually choose when to run these summarizing features myself rather then let it happen in the middle of something big, which isn’t great when it happens.
So I have been testing AntiGravity the past few days, and it is insanely good, while costing half … I am thinking about cancelling Cursor altogether.