I am subscribed to Cursor Pro. My plan renews on the 23rd of each month so I’m about a week into my new cycle. I did a big codebase refactor using Cursor last week and ended up utilizing all my credits, so I’ve now started accumulating usage-based charges.
I’d be willing to upgrade to Cursor Pro Plus but I can’t find any detail on how much usage is actually included at each level. The cursor.com Pricing page just says that Pro+ is 3x usage on all OpenAI, Claude, Gemini models but it never says what ‘x’ is for Pro. Meanwhile in my Dashboard I can see data on my usage and how much of it was included but nothing that tells me what my actual hard limits were.
hi @Brian_Cohen the number is always equivalent to AI API usage and can not be expressed other than in tokens consumed and total cost in USD.
Pro plan for $20 includes $20 worth of AI model usage with an additional bonus.
Pro+ plan for $60 includes $70 worth of AI model usage with an additional bonus.
From your Billing & Invoices as well as Usage pages in Dashboard you can see how many tokens you consumed as well as how much that was in $ included per month.
The Pro+ plan gives you more than 3x of Pro usage in comparison.
Haha, are they still measuring their plans in ~225 Sonnet 4.5 for Pro?
You won’t even be able to make 225 requests with a Pro+ subscription for Claude Sonnet 4.5.
Even in Ask mode without coding, it burns through a huge amount of tokens when analyzing a codebase.
Caching helps with that, but as far as I know it doesn’t last very long, and you’ll still end up paying a lot.
It would be more honest to advertise the subscriptions as access to Cursor with some free models + a deposit for purchasing premium model requests at API price (+20%).
Caching lasts as long as you stick with the same tab. There is a range where continuing wihtin the same tab, and leveraging the existing context, can provide better usage. With the newer summarization capabilities, which effectively compresses the context, you can ride a single tab for a good while now, with far less context usage than before (I was often riding close to 20mtok usage in certain only slightly longer lived tabs…now, I don’t generally go much beyond a few mtok, thanks to summarization.) Leveraging existing context, so long as you are still working on the same task, is often better than starting new tabs/chats frequently, as cached tokens are usually around 1/10th the cost (sometimes less) of normal tokens. The recent summarization capabilities added to Cursor have really changed the way I use it, and I rarely top 1m for a given request anymore.
I do start new chats, for different tasks, but so long as I’m on the same task, with summarization, I stick with the same chat tab so I don’t lose all the important information, knowledge, “memory” that has built up over time. The “compression” with summarization is lossy, so occasionally the model will start to lose track of some details, and you have to “remind” it, but overall, I find my general usage of Cursor is more effective and more efficient overall, when I find that balance between leveraging cached tokens, and starting new chats to avoid too much on the cached token side…while leveraging summarization to the max.
Caching has a certain TTL, and it’s pretty short — from about 1 hour up to 3 days if the session isn’t used. If you don’t touch a chat for several days or more, do you really think the provider is going to keep millions of tokens stored for every single chat? That seems unrealistic.
It also doesn’t work if you switch AI models within the same chat. The cache might work within a single provider, like the Claude or GPT family of models. But even within one provider it can become an issue — for example, when switching from Claude Sonnet 4 to Haiku.
Chat context summarization is a useful feature and often better than just a sliding window. However, it’s not entirely clear how it interacts with caching. Quite often, the AI rereads a file after summarization, even though it should have been in the cache. I need to observe this process for longer.
I also prefer to start a new chat if the AI fails at a task. Summarization tends to make the AI repeat the same mistakes or rely on old incorrect conclusions. It feels like going in circles.
I usually use a single model in a single chat. I don’t switch models mid-cht. I might use different models across different chats. As for chats that I’ve kept going for days, aside from the 8 hour nights sleep, they were kept alive by my continued usage day after day. Same tasks, though…I don’t reuse chats for unrelated tasks.