Seems like the best deal you can get right now for AI coding.
I regularly hit over $100 a month in auto use - so seams like a steal. (on the other hand it almost seems too good to be true). I’m curious what have others done?
I cancelled and went to CC Max 100, although I had a year sub so I can always renew it if something interesting happens before next April.
I’ve had the rate limit warning a few times in the early days but only like 30 mins before the reset. I’ve learned how to prompt better now I guess.
what is the difference between CC Max 100 and say…. the pro plus plan or ultra plan?
Are you paying 20 + 100?
Already had it and am keeping the annual for this too. Can always cancel later.
Sounds like even Claude Code has drawbacks where you just get rate limited. I guess it just comes down to what tool people prefer.
I did and I’m really concerned about the quality of the code lately. Anyone notice a huge decrease in quality ? its like they are only using the cheapest of the cheapest models in Auto…
1 Like
I had to stop using Auto, was a big Auto fan till a week ago … lock in auto before Cursor degrades Auto performance even further ? I dont think so
Auto has its limits. I have been using gpt-5-mini when I need something a bit more intelligent. It appears to not count towards your monthly usage limits, but that may change at some point. Regardless its like a penny per request if did count towards my usage.
What do you recommend if I paid for the yearly subscription? I understand we have lile 20$ allowance per month woth the yearly plan ? I don’t want to endup with 500$ per month. I use cursor a lot every day.
The auto model’s solving ability has degraded over the last three weeks. They can also replace the model behind the auto name with a less capable one, and you have zero guarantees about that.
I purchased Copilot, which gives you unlimited (only rate-limited) queries for a model with an actual name, like GPT-4.1. This is much clearer than having auto.
My yearly Cursor subscription ends on October 3rd, so I will fall under the paid auto model after that.
I tried to just go with the flow for as long as possible when it came to Cursors wild swings in pricing and provided service, simply because I have actual code to write and don’t have time to spend fighting other stuff.
So I got in the habit of using Auto, almost all the time, only opting for cost models when doing more advanced tasks. Actually got back to coding more directly myself, so not all bad.
Then they go “Oh, sorry, no Free Auto either” like what, a month after they told us to be cool with the new pricing because we had free Auto?
Sure I was tempted to lock in on a full year plan to maintain Free Auto, but let’s be real here, ain’t now way I’m doing.
There’s no way they are gonna give value for that, or if they are then I’m not risking it by throwing money at them as a reward for this mess.
I can see it now:
“To provide the best service possible to our users we will be optimizing the Auto mode to now focus primarily on the fastest models available. Our usage data shows that users primarily use the Auto mode for quick, simple fixes, and we estimate that transferring to models specifically tailored to this task will help further improve the user experience. These models have a context limit of 25k tokens, but are expected to be well within the needs of most common use cases. For more advanced tasks, the top-end models of our extensive model library is available and usable as part of the monthly usage quota.
Note that this change also applies to grandfathered free-Auto accounts, but still maintains the free-Use until next payment date. For other users, the optimized Auto mode comes at a new cost of 0.2$/1M tokens.”
* Models used are not disclosed, but includes Claude-Haiku-Micro and Gemini-0.5-Fast
** Auto model choice is now influenced by total Auto usage over a undisclosed period of time.
- Some forum moderator after having typed “Save me Grok”
People are probably hammering Auto hard to get the most bang of their bucks until the change, resulting in auto being absolutely garbage from whatever models they use behind the scene getting overworked. So not only do we lose the free auto, they probably wrecked functional usage of auto up until the change.
I just love unpredictability as a developer, nothing beats having stuff work significantly differently on a day to day or intraday basis. /s
1 Like
To be honest, even with Auto not being unlimited or if it becomes degraded to the point of useless. There are enough cheap/free models that Cursor provides access to that most users would not have an issue. Using GPT-5-mini or even the new grok-code which is really cheap and both are much better than Auto.
So then the real reason to stay with Cursor is you prefer the tool and workflow. The cost of requests between competitors is all going to become the same eventually I would assume. And with Cursor the cost may be cheaper if they can optimize the requests to use less tokens then if you were directly calling gpt-5 or something. $16/mo for a tool with tab autocomplete, restore points with each request and the ability to easily switch between cheap and high end models with $20+/mo credit, is worth the $16/mo imo.
And if you do have unlimited Auto locked in for a year and it is somewhat degraded, you can use it for boilerplate requests where you tell it to copy code patterns, create classes, or basic non problem solving, design requests. Or Auto may not be further degraded and that would be great. I think people keep forgetting it’s $16 for a tool, not just for $20 credit of requests. If you need a programming assistant, Cursor is by far the best based on my research. If you need a service for just feeding prompts into and not really coding yourself, then you don’t need Cursor and should just shop around to see who will give you the most requests for $$, but I bet that will constantly change as we have seen.
Also I am just over halfway through my month and I already have $102 worth of requests for $16/mo that I pay. This is kind of insane. And I have only used $10 out of my $20 of monthly usage, so I can still make gpt-5/sonnet-4 requests when I need it. I basically stretch it out by using Auto for boilerplate, gpt-5-mini for everything else, and gpt-5 only if gpt-5-mini seems to struggle or if I know for sure gpt-5-mini is not going to be able to do it. And once grok-code exits the promotional week, I will probably use that before gpt-5 since its stupid cheap and effective.
1 Like
What free models are you talking about? The Auto model was the last free one. All others that are currently free are only available for a small testing period.
With the new pricing and the same usage as in your screenshot, you will be completely blocked even from the auto model mid-month.
Gpt-5-mini is free. It’s not stated anywhere, but it does not count towards my monthly usage like the other models do when looking at the in application usage. It could change and not be free at some point, maybe after 9/15 it wont be. I also heard deepseek is free, but I barely use it so I can’t confirm it.
I would be blocked from Auto after 9/15 if I was not grandfathered into it, so yea this may only last a year for me, but by then who knows what other alternatives or the state of Cursor will be in.
Yeah I noticed the same when I used Claude Sonnet it said 100$ of APi cost, I was like “Whaaat ?”
How can they survive financially this is insane. I’ll just try GPT5 mini, I heard a lot of people saying its really the good deal. Because yeah since 2 weeks maybe, the Auto mode starts to be dangerous and really provide ■■■■ quality.
Also try grok-code, its free for a week, but also inexpensive.
1 Like
I tried it was crazy fast ! But got stuck suddenly. But really impressed of the speed and quality for the few answers I got
1 Like
Oh wow I struggled to get a proper websites with all models. Grok Code built a sick one in a few minutes this morning.
The old Cursor plan (500 requests) is still the only practical option, glad I kept it.
Subscribing to Claude Code and GPT-5 Codex CLI at $20 each seems like a solid alternative. Top models that carry you through the day, even without tab completion - but who actually types anymore?