Guys this is the first time i'm going to use AI for agentic work, coding and hading things with MCPs, which model to start with

Not true.
At $20, i'm assuming you are getting the cheapest plan right?
Cursor gives you roughly give or take 700million tokens for the month. This includes your 3rd party API usage.
I said roughly because it depends on your model, the more expensive model, the slightly less you give. But its usually around the 700 million range.
Then it gives you roughly 130million tokens on Grok bot per week. So that's an extra ~500million tokens for GrokBot standalone. These 2 pools do not mix.

First month of usage, after you finish your standard ~700million tokens, they give you something called "Slow-Mode" AUTO. Which is just their regular auto but your on the "free queue", same model, same stuff, just it may take more than 10sec before your prompt gets to their servers to be processed. longer if during peak time.
Second month, they reduce or sometimes just outright disable that "Slow-Mode" AUTO. so once you reach 100% usage (~700million tokens) it will tell you to upgrade or u cannot use it.

BEFORE elon took over, that "slow-mode" was always there, even if u subbed for 2 years.
That's probably where you are reading the "unlimited" tokens and no limit
One thing I’d be careful with is treating that ~700M number as a fixed Cursor allowance. Their current docs describe Pro as separate monthly usage pools, and the model you pick changes how fast you burn them. Once the included usage is gone, the documented options are pay-as-you-go or upgrading. I’d just start on $20 and watch the Spending tab before planning around a token number
 
If you’re just starting, I’d probably go with Cursor + Claude rather than trying to pick the “perfect” model from day one

Cursor is pretty beginner-friendly, and you can switch between models depending on the task. For agentic work and MCPs, I’d prioritize a good workflow and context management over chasing the latest model
Start small and see where you actually hit token limits before paying for a higher plan
 
I'm using a set of them all.
You can set it to dispatch multiple agents assigning to each specific models depending on the tasks

Fable for planing and brainstorming and structuring the project
Free ones for the regular sizing and data gathering
And med effort ones for actual coding.
Codex is used to judge and I use it as an external audit and review operations (sometimes it gets funny watching openai roasting anthropic)

I follow a specific handoff framework where i have a brain session orchestrating all the above.
 
One thing I’d be careful with is treating that ~700M number as a fixed Cursor allowance. Their current docs describe Pro as separate monthly usage pools, and the model you pick changes how fast you burn them. Once the included usage is gone, the documented options are pay-as-you-go or upgrading. I’d just start on $20 and watch the Spending tab before planning around a token number
what?
are you even saying.
I said give or take....................
 
Back
Top