Thank you for the detailed clarification and apology.
I have to say, that was a very generous grace period. It does seem to have been enabled recently, given that many of us were “graced” with usage worth hundreds of dollars. This brings up some critical follow-up questions about what we should expect moving forward.
For instance, could you tell us precisely what this grace period entails?
Is there a time limit (e.g., it lasts for 3 or 5 days after the $20 credit is used)?
Does it function like the old ‘slow requests’ system, where resting for a while restores some limited usage?
“You’ve saved $200 on API modal usage this month with Pro.”
These are the unanswered questions we still care deeply about. Because frankly, the current scenario where a banner boasts about a $20 plan yielding $200+ of usage is unreasonable. While it might seem like a bonus for Pro users, this practice fundamentally undermines the value of your Ultra plan and is unfair to those who pay for it.
If you could provide clear rules on how this grace period actually works, I believe the community’s confusion and doubts would be resolved. We truly appreciate the new efforts in transparency, and achieving ultimate clarity on this is mutually beneficial for both Cursor and all of your users.
We were not clear that “unlimited usage” was only for Auto and not all other models
Like it was too much of an engineering challenge to display the consumption of the “included usage”? Sorry, I don’t buy that explanation. Even now that usage information is quite hidden.
We moved from request-based pricing to $20 of included usage on Pro.
It was disappointing to find that out only when I’ve almost consumed that “included usage” because I was certain that it’s unlimited and I carelessly created big contexts when AI couldn’t produce a working code, and I just carelessly pushed it to “try harder”. Previously, when I clearly knew the 500 limit, I never got even close to using it up.
I guess I’m tired of trying to figure out things with cursor:
I’m trying to understand the value proposition here. Given that you now offer $20 worth of model usage, what’s the advantage of using Cursor over directly accessing Claude-Code or Roo-Code with API keys?
Using the model providers’ APIs directly seems to offer more transparency in terms of cost and usage. Could you clarify what benefits Cursor provides in comparison?
the pricing change appears to have occurred mid-plan. For users whose pricing model was altered during their billing cycle, please consider offering a refund or prorated adjustment. This would help maintain trust and fairness.
$20 of frontier model usage per month at API pricing
An option to purchase more frontier model usage at cost
Is this meant to be a clarification of existing terms, or a complete overhaul of the pricing structure?
What about the previously unaddressed rate limits, burst limits, and local limits? And how do the Pro+ and Ultra tiers fit into this new scheme?
This constant, arbitrary, and poorly explained modification of rules feels disrespectful to users. It’s a pattern of compounding errors, where clarity and consistency are consistently lacking.
I personally feel that this is the death of cursor.
I would rather develop some basic mcp locally (search web, search files, edit files) and use a api keys for models I like. Cursor seems very opiniated on which model should I use.
Let me analyze Cursor’s pricing history: Originally, Cursor’s calculation for slow requests was inaccurate, failing to determine when to implement this dynamic slow request mechanism. For example, after a user exhausted 500 requests, slow requests—sometimes at 1,000 requests or even 5,000 requests—were all lumped into the “slow request” category. These were painfully slow, taking about 10 minutes per request. The Cursor team couldn’t control this, leading to massive complaints. Ironically, financially, they raked in huge cash flows during this period.
Later, they revised their plan: scrapping the 500-request limit (likely after negotiating better terms with model vendors) and introducing “unlimited slow requests.” The idea was noble—dynamic request handling—but technically challenging. In reality, most users still triggered fast requests, and meanwhile, complaints poured in about unclear pricing. Strangely, this was actually the period when users benefited most: unlimited max requests surged, but paradoxically, fewer people were willing to pay $200. The massive request backlog forced Claude (the model provider) to question Cursor, and financial revenues plummeted.
To reverse this, Cursor tried technically throttling users who exceeded limits, but it didn’t work. Now, the new plan reinstates a 500-request cap for premium models while keeping the “Auto” plan unlimited. This is designed to maximize financial gains: the Auto plan will never route to Claude 4.0—they claim it “automatically adapts to the best premium requests,” but Claude 4.0 is off-limits. This has eroded user trust in Auto.
Which brings us to today: Cursor raised prices. The AI era feels like it’s regressing—all they want is bigger profits. Claude remains the undisputed king in code models. Others keep claiming to surpass it, but users vote with their feet. Here’s hoping more model vendors will actually compete in the code space, instead of just talking the talk.
Yes, it’s very likely. The story is that they tried something (unlimited) it didn’t work out, and now they are changing it in the middle of a billing cycle and telling Pro subscribers “sorry, you’ve used up all your limit”.
That’s a total break of trust for me. I’ve been using cursor over claude code because I like vscode approach more than cli/browser with long agentic background tasks. I’ve never cancelled chatgpt even when I didn’t use it much, or use their “codex” only sporadically. But that kind of attitude from Cursor is just disrespectful. I’ve just opened https://www.reddit.com/r/cursor and it’s full of the complaints about that.
So what was “unlimited agent requests” or “generous rate limits”?
You meant it was “Auto mode only” not all models.
What a silly thing. Super disappointment.
Pro plan currently covers about 225 Sonnet 4 requests, 550 Gemini requests, or 650 GPT 4.1 requests, and the vast majority of Pro users do not run out of their included usage.
But vast majority users stick with Claude 4 Sonnet, 225 requests per month is worse.
“Pro plan can cover certain amount of usage” does not MEAN you can lower the request limit threshold. You did wrong decision.
I mean, I don’t know what Cursor thinks, but legacy 500 fast requests per month is much better.
New pricing is pure garbage.
We’re now restricted and constrained more than before.
Why are your service making worse?
Why you did not make an announcement on X or via mail about new pricing? Why did you only notify on forum and documentations? Nobody reads that! you did disrespectful move.
i just put a 2 dollar limit , guess what 1.5 dollars was for 4 responses , i was shocked , can i know if i get refund , i didnt pay the 1.5dollars bill but refund of the quota usage
Is it really possible that 9B company can not hire PR, but manages to make its user base more and more #!@!#$!@ every single time it tries to fix something with another statement?
Hello World, I was using github copilot for the 10$, and github copilot was cheaper and does not include the option of auto load libraries, and cursor yes, but if you want to reach for copilot users make the pro plan cheaper too, like 10$, 20$ for international users that wants to start will be better for them, just saying, but if you user are you reading this reply if you want a change on the billing price and reply this message with “I want it”
I hope Cursor will offer a longer grace period for the current billing month for everyone, because we originally didn’t know that “burst limit refill slowly” actually means it only refills at the start of the next month. It’s quite misleading.
