I believe on non-MAX requests, the length shouldn’t matter, but for users on Pro, MAX is still billed via usage based pricing and will therefore still get affected by conversation length.
1 Like
This is still possible! If you use our standard Pro plan, and set a usage based pricing limit of $80, you’ll get the equivalent of a $100/m plan!
And if you don’t use that full budget, it’ll actually work out cheaper!
So ultra is not context limited.. and basically “MAX” all the time? That is my main concern in moving from claude max / claude code.
Ultra is not inherently MAX all the time, but comes with 20x more inclusive MAX requests than Pro.
Due to the cost, I’d almost never recommend using MAX all the time!
What is this 20x more inclusive MAX requests? Separate rate limit that resets every couple hours for MAX usage? or same rate limit counter that just gets reached faster with MAX-usage?
Can you please explain, does hitting the rate limit mean we can’t make any requests at all, or are we just put into the old slow queue?
1 Like
Look I am not even sure about the context with Claude Code, I just know for 200 bucks I get Opus 4 and the context is huge and the tool somehow automatically works. I gave it instrunctions to checkpoint each change, but I prefer the Cursor interface for chats and checkpoints. If I can get full day of work ie 10h+ from a subscription without slow pool and marked context shortcuts I will take that subscription. Right now it looks like for 200 bucks one still doesnt get the same as in claude max which is really full context for opus 4. I love the fact I don’t have to wait, I always have huge context etc. Interface is cooler with cursor, but actually one gets accustomed fast to work in CLI only if one understands the git checkpoint stuff. Checkpoints and hard resets, etc. Right now looks like Cursor is similar, but there’s still not full unbridled anthropic use for a straight month with their ultra plan. Cursor still tiers context into “Max” and non “Max” to save costs.
1 Like
yeah.. I’m pretty much in the same boat as @Decurion99
it only costs money if you don’t have claude max.
But still curious what you mean by more max requests. ( under ultra ) Max is always usage based right?
I get that when a LLM provider has their own solution, its a up hill battle for you all to make any money.
So here’s my cursor journey so far and why I cancelled. I loved the new AI coding thing and was attracted to unlimited great model use for 20 bucks. With the new sonnet and thinking stuff and the student influx the slow pool became a joke and it became clear that there’s no home for the intense users in Cursor - the majoirty is probably within 500 requests so all these moves especially the student move were great during fundraising to jack up users and projections. From an intensive user perspective claude max was the godsend all in one anthropic subscription. No more waiting for Sonnet 3.7 and 4.0, but expensive. Now I love the cursor interface and love it, but it seems the golden goose which is a ca. 99 USD subscription for daily unbridgled anthropic use is not existent yet and Cursor itself to save costs uses some tricks to compress context and auto route some people to cheaper models. For ■■■■■■■■ guys just wanting fast anthropic it is claude max still. The unclear limits to contexts, lack of guarantee on fast availability of anthropic models still puts me off. With Claude max since a few weeks I am coding away with blazing speeed, no breaks and huge context. That’s difficult to let go off for just an UI. Cursor - pls. make a great offer that matches Claude Max.
3 Likes
@Decurion99 aligned 100 percent. But I don’t think they can. Anthropic holds all the keys.
1 Like
Thanks for chiming in to provide some clarity. It’s still not 100% clear to me, though: If I choose to be on the legacy plan (500 fast requests), will I still get the regular/slow pool? How would that compare with the new heuristics for the new way requests are handled in the new plans?
@danperks I really appreciate your fielding these questions. What I can’t work out is how I can reason about usage based pricing now that the request count data has been removed from the dashboard. I really liked seeing that and it helped me to know when to add spend to my account and what I was getting for it. When I do hit rate limits I will want to enable usage based pricing–but without the visibility we had, I will have no idea how to reason about how much to add and how long it is likely to last. Could you provide some insight into this? I am very keen to drive more usage to background agents, etc. but am not ready for Ultra plan but without the visibility–seems like a non-starter.
Why would someone prefer a guaranteed $200/mo if they can just turn on usage-based billing and possibly pay less if they use less?
And if someone is consistently spending over $200/mo in usage-based pricing, why is Cursor giving them more for less?
Because Cursor can strike a much better deal when they buy bandwidth in bulk in a long term contract.
AND they have to keep the pricing low enough so that power users don’t switch over to the competition, such as Claude Code MAX.
@danperks just to confirm. Max != Ultra.
Ultra still has the same context limiting behavior of a normal request?
I don’t think that’s the case; I think statistically across all 200 usd users they’ll make a profit with the pricing from anthropic being the same. Not all 200 USD people will use all the bandwidth and they’ll use context to modulate cost. I just want to trust that I get unlimited Anthropic at full context for this price tag. No slow pool, no non availale servers, etc.
Ultra includes both standard (almost unlimited) and MAX (high allowance, ~20x the Pro plan) within the base subscription.
While not every request is MAX, you get a very high allowance of MAX usage, which when used correctly alongside non-MAX requests (for simple changes), should mean even power users shouldn’t hit rate limits.
1 Like
my monthly spend on sonnet 4 max mode would be like 3k at 10 hours a day 5 days a week if I am focused. granted I am usually not so its closer to 2. for people like me, this seems to save a ton of money. sonnet 4 is unusable without max mode and it is an expensive model. before the ultra plan, Claude code was my best bet. now I can stick with what I know. I assume a decent qty of users like me are pleased with this change!*
*I am a real person, and cursor did not pay me to make this post. wish they did though
1 Like
I think its potentially a good deal. I am confident I am spending more than 200 dollars a month in spend with claude max. (like if I paid via api key ) I’m aware on average they will probably make money.
And as we have discussed its a better UI than claude max, and I think o3 pro is a better planning model than even opus. So it would be nice to have wide access to any model.
When you avoid my points and fail to answer my actual questions, particularly concerning consumer privacy laws, and then accuse me of ‘hijacking’ the conversation, it’s quite frustrating.