3.5 sonnet(auto) is not working correctly and can’t complete our expectations. please we want to have 3.7 sonnet in agents after reaching limit too. if it is slow too it can help a lot like with us like earlier which we had.
@IsaacHopkinson
1 Like
Just use auto, it works great!
1 Like
@luke_thomas I can’t tell if you’re serious or joking.
2 Likes
Your right!!
man!!! auto are not even near to the premium models.
Why not also Sonnet 4?
any of them is ok (sonnet 3.7 or 3.4)
Yes please at least enable sonnet 3.7 in slow mode once RATE limit hit. 3.5 makes stupid changes all the time
2 Likes
I’m not joking auto works great. I have never hit a rate limit, and my projects are more complex than yours.
1 Like
All of the models make stupid changes all the time, if you are on auto, those mistakes get caught and fixed faster.
Yes because they reduce you to sonnet 3.5 at some point, of course you never see the rate limit then … And I must say this is not correct. 3.5 makes much more mistakes on complex tasks than 4.0!
2 Likes
Yea, so? Sonnet 3.5 might take a few seconds or a minute longer to solve a problem, but that’s why you multi task, and have many instances working at once and on many different projects. It doesn’t matter if it is slower, I am only limited by how fast i can type requested features. Bug testing is fully automated for me because i have very specific prompts.
auto is fine. If you disagree it is a skill issue on your part.
Well we agree that we disagree, and it’s not failure on my part nor a timing issue. Sonnet 3.5 is good for easy and medium tasks, but it fails a lot on complex tasks where Sonnet 4.0 is brilliant. If it works for you fine, great for you, but I need the high skills of sonnet 4.0 most of the time.
I promise it is a failure on your part, but you can improve with practice. I have gotten programattic image generation to work on ollama running locally on a gpu with 8gigs of vram. It doesn’t even use stable siffusion, and it can make images of things it hasn’t seen before or been trained on.
Wth are you talking about lol? What has running a local model on your GPU todo with limited performance of sonnet 3.5?
I am programming since 1999 and I could do the work all by myself, but this is not the point! I use AI as it improves my productivity and only Sonnet 4.0 can work with complex tasks without making mistakes - that’s just a fact many in this forum will confirm.
for my work auto model doesn’t work well and most of the time it just waste my time. But sonnet 3.7 is so much correct and really help me.
My point is that auto on cursor can do things people sometimes claim is impossible. It can do very complex things, making image generation work so efficiently it can run on 8 gigs of vram is not easy, and cursor did it in just a few hours.
Why does it matter if it makes a mistake or two periodically? I just run it and walk away, or start prompting another instance of cursor, and if it takes a few extra prompts to fix it’s own mistake the result is the same but it is cheaper and faster.
What work do you do that requires a specific model? Most users are perfectly fine with Auto, which is why so few people come to the forums to complain about it. There’s only like 50 people complaining. How can you know it is not a skill issue? Also auto does use sonnet 4 if it decides it is necessary, and even opus 4 makes plenty of mistakes.
I think you don’t get it. Try to edit a complex script like a document manager or analyzer using AI API, GUI, input output processing and more. Sonnet 3.5 just makes mistakes in such a complexity
Opus makes mistakes. Auto fixes it’s own mistakes. What makes a document analyzer complex? What would be complex enough for you?