Can you explain how the Ultra plan rate limits work? Specifically:
Are the rate limits based on # of tokens, # of requests, or both? Can you say exactly what these limits are? This would help me know what to optimize for.
If I use a Max model like Claude Opus, does it only consume near-term rate limits (e.g. # of tokens or requests in X recent hours) or longer term limits (in X recent days)? This would help me know if experimenting with more powerful models will unknowingly consume a month’s worth of usage.
Is there any way to see how close I am to hitting a rate limit? This would be useful.
Hi @danperks, can you please help answer these questions? I’m an Ultra subscriber and I’d like to better understand how the subscription works. Thanks!
I am also an ultra subscriber. ive never hit a rate limit. I am scared to use opus in max mode ont he chance I hit a rate limit and then never get back to normal. kinda frustrating
FYI, I never heard back on my questions here or to Cursor support so I cancelled my subscription. Especially after they charged me $34 on top of the $200/month despite not having usage-based pricing turned on and only using 764 Sonnet requests vs the estimated 4,500 requests that Ultra allows.
I know Cursor has growing pains right now and I hope they work through it. I’ll consider resubscribing then. In the meantime, I’ve bought Claude Max to use Claude Code and it’s working fairly well.
Somewhat similar situation. I was charged triple the normal costs for my pay per usage and I only did 1/3rd of the coding done as the previous month. The new pricing is a scam. I was getting charged 2-4$ per request as if I was running a premium MAX model. I switched back to the old pricing model and I am only getting charged 8 cents a prompt now. I submitted photographs, evidence, etc to Cursor and never heard back. BS
"The $34.04 charge from July 6th is for additional usage beyond your Ultra plan’s included limits in June. While the Ultra plan includes significant usage (approximately 4,500 Sonnet 4 requests per month), I can see you’ve made extensive use of Claude 4 Sonnet with thinking enabled (765 requests with 31.9M tokens), which exceeded the included amount.
You can view a detailed breakdown of your usage and charges at Cursor - The AI Code Editor (click the “Subscription Usage Summary” box to expand the view). This will show exactly which models you’re using and their associated costs."
My reply:
"1. I’ve already disabled usage-based pricing. How can I still be charged for usage?
Thank you for clarifying that Ultra means approximately 4,500 Sonnet 4 requests. How could 765 Sonnet 4 requests exceed those limits? Even with thinking enabled, I don’t understand how this can consume 4,500 requests worth of usage.
I couldn’t find the “Subscription Usage Summary” text on the Usage tab. Is this an AI assistant hallucination?"