It may be the best coder but with the new limits, you can’t get much done anyway now. It’s just for planning and fixes other models can’t make. It seems like a disaster for Anthropic. They’ve basically made their own models irrelevant.
Hey, can you clarify what exact limit you’re hitting? Are you seeing a usage limit notice in the editor, or is it something else?
On the Pro plan, you get $20/month in API usage, and models like Claude Sonnet 4.6 can use that budget faster because they cost more per request. You can check your exact usage breakdown at https://cursor.com/dashboard.
A few things to know:
- Auto mode is tracked separately and doesn’t count against your $20 API budget, so it’s a good way to stretch your usage.
- If you need more Claude specifically, you can enable on-demand usage pay as you go, or upgrade to Pro Plus $60/mo, which includes $70 of API usage.
If you’re seeing actual error messages or rate limits beyond normal usage caps, share a screenshot and we can dig in.
It’s not a Cursor issue:
It seems like coding is transitioning into something only the wealthy will be able to compete in:
Just adapt your workflow…i dont hit limits because I follow a 70/30 model…70% of the heavy lifting is done by the agent after planning and this comes after design work on Claude.ai so I end up with the right information for context loading.
If you’re gonna do waterfall development aka “build me a solution” you’re gonna burn that limit for sure but if you manage the work in an agile format break down each stage into manageable tasks that are easy to review its way more productive and I get to do the 30% the refining and refactoring, bug fixing and movement to the next stage or production.
Leverage other agents and services and you get a nice workflow.
They need to manage resources and costs (also monetise their offerings) maybe one da7 we will get local agents that can do a lot.
1 Like
Boiling Frog Syndrome :))). Don’t get too addicted to it, or they will use it against you
I adapted my workflow because there is no choice. But most of my work takes about 4x longer now. What I am saying is we may be seeing a situation where the wealthy can afford the top models and thus out-compete everyone else, profiting from the advantage leading to an even greater advantage. The cheaper end of the spectrum is getting better but may not be “good enough” for a long time, if ever, to compete on a level playing field.
One way to stretch the limits is using strict .mdc rules to keep the model on track. Most ‘limit waste’ comes from the AI hallucinating older patterns or going in circles with breaking changes (like Next.js 15 async params). If you codify the architecture in rules, you usually get the right code in one shot instead of five. I just shared a set of 22 rules specifically for Next.js 15 / Supabase hallucinations if you want to try that approach: 22 .mdc rules that prevent the most common Next.js 15 hallucinations
1 Like
I’m having no such problems. I use rules. This is basic economics. The best model available is not very available. It’s not a me issue.
Upgraded my plan to the pro, after one month, and the rate usage is flying. 2 days and already at 50% API limit and 20% of my auto.
I was using Claude, Gemini 2.5 flash. Even tried GPT 5.2. detail task list.
I was using Claude code for the planning, so majority of the heavy lifting was already done.
But the usage credit keeps jumping. This might be a Claude issue.
Honestly auto mode can only complete very very simple task, but half the time will break something else.
I used to get at least 10 days out of cursor API, without using Claude Code. And at least 20 days from cursor.
What’s happening with the rate limits? Did it get updated again recently?
As a monetary economist, I think it is relatively easy to predict that AI tokens will eventually become money- the most saleable/marketable good, and that all other goods will eventually be priced in tokens. The world is going to be unrecognizable in 10 years.
1 Like
I get why it feels that way, but I don’t think it makes it useless. It just changed how you have to use it. If you try to do big, all-in-one tasks, you hit limits fast. But for breaking things down, reviewing code, or handling tricky logic, it’s still really strong. It’s a bit less convenient, but the output is usually more consistent that way.
I am pretty sure it is still getting worse. I was able to use significantly more tokens yesterday with Claude than today.
It looks like about a 50% reduction in the allowance for Claude before hitting rate limits from yesterday to today.
@neverinfamous Exactly same here, from yesterday to today Claude became useless. I ate up my usage limits within 1 hour or so, and I am on the pro plan for Claude.code, so with Cursor I can do the same, but then must switch to other models. Cursor’s biggest issue is its hard limit, and then you have to wait for the rest of the month.
1 Like
Yes, it surely presages a dramatic rise in price, the inevitable outcome when there is greater demand than supply.