Thanks for the feedback. I had hit my limit and was using gpt-5-mini and was still surprised what it could do, but the ones you mentioned may even be better, so going to give them a try. It’s good to have a couple very inexpensive models to use as your daily driver before moving to sonnet or opus.
Also in response to your previous post, I think inexperienced developers are going to get in this AI trap of spending more and more money because they don’t understand the underlying code enough to discern if they should be using a low or high functioning model. Instead they are defaulting on using the most expensive models out of fear of a mistake being made, when it’d be obvious to any experienced developer based on the context and nature of a request (and if they are reviewing the changes) if the agent is even affecting a system that has the potential for creating a vulnerability. Unfortunately, it looks like these people are doomed to be paying a ton of money because they are at the mercy of Cursor, OpenAI, and Anthropic. But this has always been known with the rise of vibe coders, but now we are seeing the honeymoon phase of coding agents end and prices are increasing (and I assume will continue to), since now so many people have already adopted it and can’t go back.
And I think this is why Cursor introduced the “Premium” feature, so inexperienced users can have Cursor determine which model is best based on the request.
I would just say TOUCH’E to all of you! yes I loss you won, you are the master mind coders, you are the billionaires which I know you are not, you still didn’t even try to understand a bit and I’m 100% sure its the cursor team behind these accounts, so yeah I cannot win with you so good luck and don’t bother anymore…!
And no we are not Cursor employees, but I just noticed that @Artemonim and I are 2 of the only 4 non cursor employees on the entire forum that have the “Respected” badge. So you were getting prime help here. Good luck.
This is a great example of why choosing different models matters. The spread here is insane. If you were only using sonnet thinking you’d be in the thousands cumulatively.
Cursor Settings > General > Rules for AI is not here in:
Version: 2.6.22 (user setup)
VSCode Version: 1.105.1
Commit: c6285feaba0ad62603f7c22e72f0a170dc8415a0
Date: 2026-03-27T15:59:31.561Z
Build Type: Stable
Release Track: Default
Electron: 39.8.1
Chromium: 142.0.7444.265
Node.js: 22.22.1
V8: 14.2.231.22-electron.0
OS: Windows_NT x64 10.0.26200
I put a rule in but does not follow the rule to never use subagents
1 chat request is now costing me 5x the price. This has to stop immediately of course totally ridiculous, subagents are not saving money, they are generating more request and the money goes one way…> straight into Cursor. Yes I’m not amused by this behaviour change of Cursor
Hey @Someguy, I see the screenshot. You have claude-4.6-opus-high as the main agent plus two composer-2-fast passes, which are the subagents. On legacy Pro with request-based billing, each subagent counts as a separate request, so one chat can grow to 5+ requests. That’s by design for that plan, not a bug, but I get that it looks bad.
The path General > Rules for AI from my earlier message is outdated. In 2.6.22, add the rule like this:
Cursor Settings > Rules > User Rules (or the file .cursor/rules/no-subagents.md in the project)
Content:
Don't run subagents. Do all work in the main agent context.
A User Rule is not a hard block. The model usually follows it, but not always. If you want it to be more reliable, add two more steps:
In the model dropdown in the chat, turn off Auto and Use multiple models, then pick one model explicitly.
Settings > Models: turn off composer-2. Downside: you might sometimes see tool failed if the agent still tries to delegate, so this is more of a fallback option.
