Pricing feedback/clarity - from actual users

Discussions
, ,
1

Hi, we are going to transition from legacy pricing (500 request / month) to the API pricing next month and wanted some feedback and clarity from the community. My discussion with Cursor team has only confused me than clarify.

We are considering the Teams annual premium plan ($96/user/month billed annually)

Let us take examples of 3 users and let me know the expected usage for each of them

  1. User 1 - Only using Opus 4.8 max thinking for entire month

  2. User 2 - Only using Composer 2.5 for entire month

  3. User 3 - Only using Auto for entire month

In all cases assume the below distribution of tokens

  • Input - 3%

  • Output - 1%

  • Cache Write - 6%

  • Cache Read - 90%

Can you tell me the approximate tokens each of the above 3 users would get?

I am told there are 2 usage pools to draw from (Composer + Auto) and Third-party. We are heavy Claude model users, would we able to use all our budget/quota with just Claude or we will be compelled to use a part of the budget/quota on Composer and Auto? My discussions with the support staff/bot have led me to understand the ratio of allocation is not disclosed or guaranteed. Hence, it appears that Cursor reserves the right to direct 90% of the budget/quota to Composer models and only allow using 10% budget/quota for Claude. Is that right understanding? That is concerning to me, what are your thoughts?

If I pay a store (that sells blue and red candies) $100 and I like blue candies, I would like $100 of blue candies. It appears that I can only have $70 of blue candies at max, hence I am compelled to buy $30 of red candies even if I don’t like them? What do you think? Also, it is not clear if I will get a total of $100 candies or less or more.

Why is Cursor capping how much Claude or GPT models should I use when those are already priced at API + CTF? Feels anti-competitive and promoting Composer models.

4

In fact, I don’t recommend switching your billing method because API-based billing will be very expensive. If you use Claude Opus 4.8, 500 requests could easily rack up nearly a thousand dollars in API fees—or even tens of thousands if every task is heavy-duty. Once you switch, you’ll be limited to a $100 credit. However, if you only use Composer 2.5, it doesn’t really matter whether you switch or not; that model tends to be lazy and often stops mid-task. Plus, the pricing for that model itself isn’t expensive. The same logic applies if you only use Auto mode; its intelligence level isn’t very high, so it tends to slack off. My thought is that if you only use Claude Opus 4.8, don’t switch. Try to make every conversation task as massive as possible to get the maximum value. There’s a reason they are pushing users toward the Composer and Auto models—primarily because they are now an AI model provider themselves. If users mainly use their proprietary models, it’s definitely more cost-effective and profitable for them.

7

Ovressa, thanks for the inputs. That is exactly where we are, we are heavy Opus users and we do not want to switch to API-based billing.

My understanding is that there is no option to stay with per-request pricing and at the time of renewal the switch to new pricing will be forced.