Why is the DeepSeek R1 and gemini-2.0-flash model, which is so cheap, charged the same as Claude 3.7?
I used DeepSeek R1 and gemini-2.0-flash, both deducted premium requests, and using @Web added one more. What is the deduction logic? The previous version did not deduct fees like this!
Perhaps you would want to read docu?
and
It specifically mentions R1
So DeepSeek R1 and Claude 3.7 sonnet are the same price, and in version 0.45, they are the same price as Claude 3.7 sonnet Thinking.
Things cost money. If Cursor can’t provide a quality at lower price then they have to increase price. It would make sense right?
Cursor also communicated why they increased 3.7 thinking pricing. That they integrated it quick and incomplete in an very old version of cursor (we are now at 0.46.8 in standard and 0.48.1 in beta) had to be fixed later on. Those improvements are noticeable
Hey, the reason this is considered a premium model is that we run it not from the official DeepSeek API but through our partner provider Fireworks, whose API for this model costs significantly more than the official one.
Where are the charges for web search and other tools reflected?
They are covered in the request under all models except MAX
Its a pointless question really mate because they never work to the same standard for more than a few days or a new release, because they just dumb down the last to make the next look better!! The proof is completely in the pudding!
I disagree, for me it works great. just did a lot of work today.
I have last couple days! Today! Its in dumbo self-destructo mode again!!
I switched to 3.5 honestly 
works like a clockwork
Not really the point though is it. But i will try 3.5 and try to repair some of the damage 3.7 has done today!!
I am keeping every bit of evidence now because when there is a class action lawsuit against these guys I will be prepared!!
Does an @Web request count as a Premium request?
Why keep worrying over endless worries with cursor. i use aug____nt. Cant write it out. The agent mode is not erratic and overzealous like cursor. Built to handle code base greater than 10k files. I tested it on a prod codebase of 4600 files and it was able to make surgical changes with high quality output without suggesting garbage code and trying to overwrite working code despite very robust rules . This tool works like a smart and loyal dog. Gets the job done. It made me cancel my cursor subscription. And no it is not winds…f or cline or all those clickbaits. This is the real deal built by engineers for engineers. I cant write the name else some triggered kid would flag my post.
Yup… careful with words they might delete your comment
Whatever, it doesn’t matter, good things will be known worldwide
