Alright, let’s get this straight—huge shoutout to the Cursor team for once actually delivering a real working Auto Mode. For the first time, this feature isn’t nerfed to hell—you actually get useful context, and you can even see how much context you have at any given moment right inside Auto. That’s proper engineering, finally.
But the absolute best part? Anyone who bought a subscription before your new pricing—yes, even students—gets unlimited Auto. Unlimited. No restrictions. No hidden throttling, no “not available in your region.” You actually kept your word for once and didn’t disadvantage the OG userbase. That’s rare as hell in SaaS, and you deserve credit because it’s pretty much a unicorn move in this industry right now.
Of course, I have no idea how long this will last with how unstable your pricing and feature sets usually are, because honestly, you people change your mind more often than my code ends up in the garbage. Still, fact is: right now, as it stands, this is absolutely awesome—so, thanks. Legit.
P.S. Don’t get comfortable. I’ll be back complaining when it breaks. But for now? Respect, seriously.
Also, let me add this: in Auto Mode, you guys don’t shove in cheap models like DeepSeek, Kimi, or Qwen (even though that would’ve been easy and dirt cheap for Cursor). Nope—you give us genuinely premium engines like GPT-5 and Sonnet. Major respect for not cutting corners and actually delivering the best to your users. Keep it up!
Thank you kindly for your feedback, we do really appreciate it and I will share this with the team. You correctly stated the reason we are keeping better models in Auto mode and improving routing as well as handling, to have it perform well as a default selection.
Anything that we could improve?
For bugs please create Bug Report as we are tracking them differently than discussions.
Surprisingly, Auto mode works really well and for now doesn’t have limits for annual subscribers. I think now that it has become a paid feature, Cursor will likely develop it further as a concept.
I’ve seen SaaS platforms built entirely around automatic model selection based on the task. This makes it possible to manage costs and bulk provider limits without losing quality. Now Cursor has this too.
At the moment, I’m mostly seeing GPT-4o, GPT-5, and Sonnet 4, with Sonnet 4.5 showing up only rarely.
I’d like to join in thanking the Cursor team for making this possible, and I hope they continue moving forward in the same direction.
