Hey @DjWarmonger,
Just want to respond to this point.
Auto is priced as follows:
- Input + Cache Write: $1.25 per 1M tokens
- Output: $6.00 per 1M tokens
- Cache Read: $0.25 per 1M tokens
Compared to many models (especially “frontier” models like Opus, Sonnet, Codex) Auto actually comes in under, save for a few smaller models like GPT-5 Mini, Gemini Flash, and Haiku.
Grok gone and now Auto gone? Hell. There is no “free” options now. Is there a point in using Cursor now? Maybe beside times when it promote some models for free?
sigh
Well i guess Qwen Coder is still free in OpenRouter. And GLM 4.7 Flash have good timing to appear.
This has been deeply frustrating and it says a lot about the leadership behind Cursor.
Before September of last year, Cursor made a bold and smart move. They reduced AI costs and pushed agentic coding to the forefront. That decision attracted many of us. It felt like we were getting in early on something transformative. A lot of developers moved over specifically because of that positioning and the promise of unlimited Auto.
Once Cursor captured meaningful market share, they introduced what they called a new pricing structure. We were told that after September 15, 2025, on the next billing cycle, Auto would no longer be unlimited. I personally confirmed this with a sales representative. They also confirmed that if I paid for a yearly membership, I would be grandfathered in under the original unlimited terms until September 2026.
Many of us took that offer. We paid upfront in good faith based on that representation.
Now, instead of honoring the spirit of that agreement, Cursor is throttling bandwidth in a way that was never disclosed and was never part of the original deal. Yes, terms of service can change. Every company reserves that right. But there is a difference between adjusting pricing and fundamentally altering the performance and value of what customers already paid for. When the product becomes materially slower or restricted, that is not a minor adjustment. It changes the core experience.
At the moment, I can type faster than the agent runs. That defeats the entire point of agentic coding. The frustration is not just about speed. It is about trust. When customers commit annually based on specific assurances, those assurances matter.
This feels like a classic bait and switch. It may make short term financial sense on paper, but it erodes long term credibility. Greed is rarely a good business strategy when your product depends on developer loyalty and word of mouth.
Many of us joined because we believed in the vision. What happens next will determine whether that belief was misplaced.
quick update, cursor stands behind their agreement and are not throttling legacy accounts. This was great news
Did they confirm that or are you no longer seeing auto slow down?
They confirmed in writing, they checked with engineering and billing to confirm same.
unlimited auto has consistently worked for me; on an older annual plan. Very thankful that cursor has honored this!
In the past when auto was not responsive it seemed like it was a load issue. Most likely to happen when a major region wakes up (for me it was Europeans haha)
did cursor have a lobotomy recently? ive been having a hell of a time getting basic things done with the auto agent…
I’ve noticed the auto mode qualty has severly become so degraded to the point the model isnt understanding the prompts correct and doing the oppersite. In realty for those who purchased pro annual before september, should still have access to the auto and premium. cause in effect its annoying a lot of people.
Auto may be downgraded, but its hard to say. The introduction of Premium is not really comparable to Auto, it is just an auto model selector where you still pay the model’s API rates, instead of Auto which charges a flat rate for whatever model it routes to behind the scene.
So adding Premium is not really taking away from Auto, but doesn’t mean they aren’t changing Auto behind the scenes. It’s fair for people to be suspicious.
I’m in the same boat as a long-time user. What bothers me most is not just the higher price, but the fact that many of us subscribed based on a very different understanding of what Auto mode was supposed to be.
Early on, Auto felt like one of Cursor’s biggest selling points. The expectation was simple: let Cursor choose for us, accept that slower responses might be part of the tradeoff, and in return get a reliable experience we could actually build our workflow around. That was a huge reason many of us stayed on Pro.
Now it no longer feels like the same product. The limits feel different, the quality feels less consistent, and Auto often feels like a black box. When it starts misunderstanding prompts, making worse coding decisions, or behaving noticeably weaker than before, users naturally start wondering whether they are being routed to lower-quality models behind the scenes. The biggest problem is that we are left guessing.
That loss of trust is the real issue. If pricing changes, then communication and transparency need to improve as well. Loyal users should not feel like they subscribed under one value proposition and later ended up with another.
I’m not against paying for a good product. I’m against paying more while getting less clarity, less predictability, and less confidence in the core feature that originally justified the subscription.
I am on the grandfathered, unlimited plan and I have not noticed any degradation or limitation in the use of Auto.
I have been coding for many hours (up to 8 hours) a day several days a week for many weeks, using Auto, and have never hit any limits or issues.
I’m very happy with it, and also very sad that unlimited Auto will end for me in September 
samesies…. but the cost of AI will be drastically different by then… i have noticed that after some hard coding with auth you get this new “taking longer than expected” thing which is very likely their way of throttling us
not for Auto until you renew.
Its coincidental that ive noticed massive agent degredation and now this pops up… Cursor hid this post so im resurfacing it, apparently you cant be too critical of things here…
But hey, just make use of Cursor’s $20 subscription for 2–3 weeks (5–6 hours a day), then configure the DeepSeek 4 Flash model and finish the job during the remaining week (for about $5 extra). Awesome!
