I love Cursor but their competitor is charging 1.25 for Claude thinking 3.7 (recently decreased from 1.5), at $15/month. It’s not even close when it comes to this. Customers are allowed to complain about price increases…
5 Likes
My requests still only cost 1? yeah I’m never going to update lol..
No worries, I’m happy to pay for excellence. I’ve been tackling repo errors and issues affecting me with a two-step approach. First, I use ‘Ask’ mode to pinpoint the core problems and sketch out a solution strategy. Then, I switch to ‘Edit’ mode for a second pass—both powered by Sonnet 3.7 reasoning. The result? 100% clean solutions. I’m submitting fixes to a massive, complex repository that isn’t even mine. So, I’m probably burning through 4 requests per go.
For this particular mega-repo, I’ve introduced a new rule to rein in 3.7’s tendencies: the ‘Simplicity Principle.’ Here’s what it’s about:
The Simplicity Principle
-
This is an active repository with tons of users.
-
Every code change must have minimal impact on the rest of the codebase.
-
Always boil solutions down to their least disruptive form.
-
Complexity is the enemy we’re dodging at all costs."
2 Likes
“We’ve made a few improvements” → Charge more
“More improvements will be rolling out”-> NO THANK YOU
It did not last =)
Well, praised be V3.1 and Cline \o/ Just waiting for a really good autocomplete
I wouldn’t mind paying double if it actually gave better results. Add an extra “premium” plan for $50/month which allows models with max context when required. For most simple tasks it can use a lower context and standard model, but when I start refactoring a large script, it’ll sometimes go off the rails, burn through a bunch of credits and leave no other option than to revert, regardless if I use/provide a memorybank, cursorrules, scratchpad with tasks and clear instructions.
Monolithic scripts are always going to be problematic. Work towards traditional programming conventions, by converting to modular coding practises. I also recommend Spending lots of time tuning your rules, take extra time to craft prompts.
The new cursor rules allow for very precise guard rails. I just spent have a day working on my rules, and actually let sonnet 3.7 thinking in edit mode help me perfect them. The models are highly aware of the rules, in spite of what some people say. You do have to make your rules organized and easy to interpret, concise.
Giant context models should not be an excuse to believe traditional fundamental coding principles can be ignored
1 Like
LOL. I had a massive rant with the support to land on this, and they mentioned it’s in the docs. haha.. I see now it’s actually the changelogs. So, screw old users right?
Cursor, good think you are the best at doing this. Highly doubt developer community will suffer this for long though.