was this a CLI tool, IDE or IDE plugin?
It is an extension tool that is light years ahead of both winds… and cursor. It starts with aug… sorry for not mentioning the full name . Cursor mods can be very trigger happy with deleting post or user accounts if u mention competing product. Just do some googling. It support unlimited chats. And unlike cursor that u need to keep creating new chat session. You dont need to worry about that, that tools context awareness is insane. That tool has an agent mode that does TDD out of the box delivering high quality code output. It doesn’t go crazy like cursor agent mode, it sticks to task at hand implements the task and generates unit test as well , so it won’t commit changes unless all test have passed. This is without any custom rules. My jaw dropped when it fixed a bug cursor couldnt fix after 20 tries.
Hint please, what alphabet does it start with , i’m also using one that no one knows about much but it’s really good
ah ok , i think we are using the same one, lets not talk about it haha , gonna get banned. You’re right the context aware is crazy i literally have like a repo with 10-20 projects and ■■■■ lots of files since ancient time, it could find the file for me lol. I think if people start flooding there , they probably will have to start nerfing it too
1 Like
Huhuhu , smart man. Now just so u know go to the discord now , they have an agent mode now that is mind boghling, its accuracy and power blows the chat away. Unlike cursor agent mode that is erratic, that one focuses on task doesnt delete code and does tdd out of the box. Its in alpha but works like its in GA state already. Lol 
agm
They will nerf @ugmet0 too once they hit critical mass. But for now its great
TLDR of this thread: MAX should only be used in ask mode. In that mode, it’s only 25% more expensive than regular Claude.
2 Likes
Did anyone notice Cursor just doubled the cost of using Claude 3.7? Each action now takes two points instead of one. WTF guys? Maybe some increase in price it is needed because you have to make a profit, but did you need to double our cost and throw out the new Max 3.7 for a lot of monetization on top of that? With the 200 tool limit, one action could cost you $10.05, with $10 in tool costs alone??
In Max 3.7, it is the same LLM, just a larger context window, which likely gets you more tool calls. It can look at more code at once, and I suppose it could use more compute, but I would think that it is simply a setting in Cursor alone, and you cover the cost with action points.
Please help me understand how you can justify doubling our price and offering a Max when the LLM model itself is no more powerful.
1 Like
No mention on how you double the price of using Claude in the regular plan?
1 Like
I really don’t want to say this, and I truly love Cursor, but I ask the Cursor team to please try using WindSf just once. Then, please improve Cursor. WindSf never causes situations like this — in Sonet 3.7, editing is never interrupted or left incomplete. It is always completed no matter what.
Not only MAX, but improvements to Sonet 3.7 are also necessary.
I am currently using both Cursor and WindS**f, and still, I remain a passionate fan of Cursor.
1 Like
It seems that the cursor can earn more by constantly making mistakes. This is a game where one plays both the athlete and the referee. Ever since max was released, it feels like my regular Sonnet 3.7 context window has gotten even shorter, and tool call errors have become more frequent.
3 Likes
With the latest updates, I’ve noticed that the regular Cursor models like 3.5 and 3.7 aren’t performing as well as they used to. On the other hand, MAX is working impressively well. It almost feels like Cursor might be intentionally shifting focus to make MAX stand out by dialing down the performance of the other models. A bit disappointed with the results.
Do you see that also on 0.48? For me it works much better now on 3.5/3.7. But i have also adjusted my prompts and rules.
There must be (meaning we users are requiring it) - a “LINTER_DISCOUNT”
Meaning we need @cursor to de_lint, and not charge for such.
This is the way.
I’ve switched to Claude Max as soon as it was available. Shortly I was getting $60 / day bills, didn’t check the pricing properly only until I wasn’t really happy with its performance. Turns out all of it was for tool calls. Yesterday I switched to normal Gemini latest, solved the issue Max struggled with quicker.
I don’t get the pricing really. There is no advantage of MAX models in genetic usage, especially considering the pricing.
It is even worse than that for me:
Max is destroying things I never asked it to touch on the subsequent requests, so in a way, MAX agent 3.7 performs worse than regular 3.7 Ask mode for 20 times the cost.
Claude is more trigger happy than Gemini, but Gemini requires more input so 
With proper test coverage and prompts I prefer Claude, but MAX does not make sense as it currently is.
1 Like
I set a $20 bill and it reach within a day – I realized the Cursor somehow rebuild the code 2-3 times.
1 Like
If people spend more requests planning and designing than actual coding then people will be a lot more satisfied. Using AI does not exempt you from being the lead analyst / project manager.
2 Likes
