My whole team was on cursor for a few months. I enjoyed using it and thought it was the most complete of the agentic coding tools I tried. The thing that got me was the cost. I was switching between Opus and GPT 5.x and was spending anywhere between $500-1000/month. I was using a relatively normal workflow, paste in ticket, plan, execute with dumb sub agents, have the ai test and competing model to validate. The business got uncomfortable with the cost when everyone started doing the same so they switched us to Claude code since it has better cost controls. So far it looks like we won't even touch the $100/month plan and some people would be ok on the $20 plan. Anthropics usage limits is a consistent source of complaint on here but I've found them to be moderately generous in comparison to cursor. Cursor also charges a $.25Mtok premium for 'routing' no matter what model you choose. 5% increase for frontier models but when you're using haiku on sub agents that's a 50% cost increase. Composer is solid but if you don't have deep pockets it's the only feasible model on their platform because of how they bill it. Being an all in one editor/agent is nice but if you're in a language like c# or Java you're already swapping back and forth with a real IDE anyway.
More precisely -> the competitors heavily subsidize which causes cursor to feel expensive.
> Recently, we purchased one of each Anthropic/OpenAI subscription plan and randomly ran long horizon coding tasks until we exhausted the weekly limit. It's widely believed that a $200/month plan maxes out at ~$2000/month worth of tokens (assuming API pricing). However, we found that the subscriptions are actually far more generous. (2/4)
Our experience has been the direct opposite. The way to get the most out of cursor was to leave it on auto - we saw it average around 3.9 cents per request under the old contract (per-seat pricing) and more like 39 cents per request under the new (single pooled cost for everyone). Composer came in a little higher, more like 50c/r, while the claude models were up into the dollars. Meanwhile, if you use Claude-the-app, there is _no_ cheap model and the default switched to Opus in April, resulting in increased costs across the board.
We have both Claude and Cursor here, as well as agents running GPT, things in AWS Bedrock, etc and its my team handle the bills...when people use Cursor on auto, costs are under control, but there's always a dozen or so whale users who'll switch models manually and blow through the budget like it's not there.
Another thing: "better cost controls". There was for example no way for us to disable Fable in Claude, but we could in Cursor. Again, the opposite experience.
For the kind of ML work I do Cursor's "Auto" was too unreliable. It would frequently try to solve relatively difficult tasks using Composer, when it should have been routing to a true frontier model. Annoyingly, doesn't tell you what model it's decided your task deserves, so I often wasted time working through a problem with it, only to realise I was on a dumb model. Then I revert all the work, switch to an expensive frontier model and pay expensive API prices to actually solve the task.
All that is to say, model selection is the main control we have over quality. Giving it up in the name of cost saving will bite you in the long run, especially when Claude Code still has such good plan pricing.
I also had a problem with Cursor auto switching on fast mode which I would never use otherwise.
Cursor really needs a stronger model than Composer 2 which is the root of the problem. Hopefully they will put out something better now that they have better datacenter access with SpaceX.
Well, auto can produce garbage code. I'd rather be able to select the model manually myself and so far it's much more economical to give people Claude Code and Codex subscriptions as those are subsidized per seat plans.
Yay! Registrations to https://merrilin.ai are open and you can enjoy a free account for as long as I can afford to give it (I have no funding lol). The android app is ready for closed testing too, if you have an Android device, mail me at mail [at] stonecharioteer.com and I'll add you to the beta testing. ios is still a pain.
Why don't you do sign in with openrouter oauth, then users can create an account and assign a key to your app, with spending limits. It's trivial, plus if you get a decent amount of users your app will appear on the leader board and that's free marketing for your project.
That's a good point, but our harness right now is very fine tuned towards kimi2.6, it's something we want to improve but it's just 2 broke dudes working on it right now.
That’s really cool! I’ve been thinking about how LLMs could help me with long running series. I read lots of stuff that’s thousands of pages long with a huge cast of characters and locations. Amazons X-Ray is okay ish but (1) not supported everywhere and (2) amazon.
It would be really fun to have a progress-aware AI that can give me a quick definition of entities like people and places. The other thing that would help is details about fictional mechanics. How exactly does FTL work in this universe? What were all the cultivation stages? I don’t need help with reading comprehension, I need a better way to flip back through everything and surface the key detail that was mentioned one time 500 pages ago.
Also, unfortunately my library lives in Kindle. Help me get it out, at least the DRM free stuff. I also use Royal Road extensively and pay for that. Would be great to have those live serials supported somehow.
You're in luck. I built this especially to target series and you can ask questions across books whether they're in a series or not. I wanted to be able to ask questions of books that aren't even related to each other together.
Royal Road looks cool. Do you know if you can extract them into an epub? I have a user who extensively reads Web Novels and I wrote a blog post about that just today. He converts them to epubs AFAIK and uses them.
Probably. But it would be at least somewhat thought-out and apply to all the AI providers. Not just the one currently disfavored by Captain Dipshit and the Sycophants.
I really don't know why business cozies up to Trump so much, given how unbelievably unreliable and mercurial he is about...everything.
I'm sure it's also a step towards requiring id and limiting access for us plebians to real power and keeping it for maintaining or growing power of those in charge. It's all an excuse to give us a Westworld season 3. Probably a better example out there..
I've found the opposite. Granted I use sub agents heavily but I've had it run for hours with far fewer tokens used than when I was previously using opus4.6-8.
reply