As on the Website stated :GPT-4, GPT-4o, and Claude 3.5 Sonnet are counted as premium models.
Why do i get charged for it when using Deepseek? I checked my fast request on my account and used only deepseek for a while now and i can see that get charged for it and my fast requests are going up.
I am paying customer and only use the premium models when i really need it as 500 fast request is a joke, so i switch to deepseek and you still charge me for it. Can you explain that?
Hey, you can check the prices for all models here. The DeepSeek R1 is a premium model.
then update your landingpage
I have looked many times for definitive answers about model costs / which models work in what modes, but this is the first time I’ve seen anything approaching a useful list. This:
Needs to be more prominent on Cursor’s main site. MUCH more prominent.
Needs to include all models that are in the program. There is no mention of the Gemini models, for example. How many people know this is a viable free option?
Side note, but it definitely relates to this table: There’s been contradictory information about how many requests are consumed by Sonnet 3.7 - Thinking. Some of it has come from people who are apparently staff (or at least have elevated badges). Some have suggested that 3.7 Thinking takes two fast requests. Some say one. Some say possibly many, and it just varies.
My personal experience so far: 3.7 always consumes at least one fast request, but it will sometimes eat two, for reasons which so far elude me. The timing is usually not instant; the second request is consumed at some point when I am still getting the answer. But it’s variable enough that I can’t find the pattern yet.
Obviously this is uncool. I understand that Cursor has very quickly added 3.7, which is nice, but the bizarre inconsistencies of how many requests it uses must be fixed or explained. This gets at the very core of commerce 101.
Regardless, I’m glad I have now seen a table suggesting what it is “supposed” to cost, which I had not seen before.
I just cancelled my subscription mostly due to that. There are alternatives.
Hi, we’ve updated the pricing faq to make it more clear that sonnet/gpt-4 are examples of premium models but not the only premium models. We’ve also added a link to the models page from the FAQ.
deepseek-v3 is free, but r1 is 1 premium request.
I thought DeepSeek R1 was supposed to blow the competition’s pricing up and make it cheaper (which it is), so why is it the same pricing as 03-mini-high/Sonnet 3.5/3.7?
Only to say, totally absurd. as mentioned in ds’s twitter, ds could be incredibly cheap while maintaining high performance
Cursor has chosen to use a Deepseek R1 provider that is 5.5x as expensive as Deepseek’s own most expensive pricing (cache miss during peak cost time period) - nearly 86x more expensive than DeepSeek’s cheapest possible pricing circumstances (cache hit during discount time period).
That’s not decision making that’s consistent with the interests of the clients.
the real question is: why is it more expensive? it’s the same model, the only real difference is it’s being served from elsewhere
Not true, there is a huge difference as the original provider uses user API prompts for training, where the provider Cursor uses does not do that.
sure thing, but compute time should be about the same, still don’t see why it should cost them x5 lol
imho deepseek r1 should be 1/3 of premium request at most, but that’s my point of view
At the time there was a ridiculous discussion here on the forum about some customers not wanting to use the pattern and
Kkkk because your millionaire and revolutionary code would go to China kkkkk so they voted and now we all pay for it