So long story short i use sonnet 4.5 for my work and becuase they have changed the pricing to be based on model usage instead of number of request that was 500 previously, i wonder how are the people with very large industrial grade projects able to use this because its charging like 1-2$ per request which is insane. or am i missing something here?
I see it as this.. There are two paths.. 1) You use expensive agents, counting on the agent intelligence to manage any contextual ambiguity you as a human might cause due to a lack of knowledge or understanding. These agents are sometimes faster to solve a problem.. but you pay accordingly. 2) You focus on using inexpensive agents, counting on your ability as a informed and knowledgeable user to make effective , clear prompts the less intelligent agents can still leverage. It might take longer to get to final code.. but it costs a lot less in terms of usage. Having a good development diary mechanism makes #2 truly viable as only then do you have the necessary context to keep the less intelligent agents in line…
1 Like
The Ultra Plan. It is paid for by the company I work for, including any token-cost overage. Most months I go over in usage, as do most of my team members. It probably costs $300-1000/mo for each team member. Some are very high throughput, although there is often a moderate amount of rework that has to be done (it is interesting, the implications of explicitly aiming for the 80% all the time, when you are a startup, though! There are definite benefits!)
For the company, the cost of paying for Cursor Ultra offsets the cost of bringing on new team members as often as we would otherwise have had to. We are a team of, I guess at this point seven senior software engineers. If not for agentic coding (and FWIW, not all of us are paying for Cursor, several are using Claude Code MAX), the team would probably have to be triple in size if not more.
So, consider that. Lets just say its $500/mo/user on average. That is $42,000/year. Now consider the cost of just doubling the size of our engineering team. Let alone tripling it or more… Its the difference between maybe a million to two million a year….and $42,000/yr. Its a no-brainer.
Even as a one-man shop…why in the world would you consider paying token-cost with a $20/mo plan where you burn through the plan-included in a few days? Ultra is the easy answer. You spend $200/mo for $400/mo usage. Or Claude Code MAX, or whatever your preference is. We were really burning tokens, to the tune of several billion a month per team member, during the summer. We were effectively a brand new startup back then. We have established code bases now, so the token burn isn’t as high. On top of that, we’ve also all been learning how to be more efficient with our token usage. So token usage has dropped. Token costs have improved a bit. Model speed has generally improved. I myself, probably go over $50-100/mo so it generally costs me about $300/mo or thereabouts, which is $3600/year. If I was working on my own project….seriously…$3600/year! That is nothing! If you managed to stay within the $400/mo usage on the base $200/mo of the Ultra plan, then its $2400/year.
You have to contrast the cost of using an agent to help YOU build your product…and the cost of hiring REAL PEOPLE to build your product for you. ONE person, assuming you paid them at the poverty level (~$32,000/year), is easily 10x more costly. One person at a reasonable tech salary is 20-50x more costly.
It doesn’t have to be Cursor, and Cursor has plenty of problems right now with their bum-rush to shove out as many half-arsed features as they possibly can in the shortest time interval possible. Claude Code MAX or any one of the other options out there, for a couple hundred bucks a month, if it is really helping you develop a viable product ON YOUR OWN? Worth it. No contest.
Not when you can build tools or products that bring in 50-100$ for each request.
Or otherwise, you can keep the fixed revenue or income and reduce costs. Paying 0.01$ to Grok to manage a bundle of texts in 10 seconds that would take you several minutes is worth it. Currently coding, research or text processing jobs are 40-50X faster in the same unit of time.
You know you can use Grok for 1 cent, but you also know you will use 100 times in a row, and it will not give you what Opus brings in one query, so you pay up obviously. Because you know it’s worth it. If not now, maybe in the future with additional experience and built-up tooling over the usage.
thank you soo much for taking the time to explain in such a detail, i guess i would need to tell my client to either hire someone else too or otherwise buy the max plan if possible because this is very understandable that a new dev will cost him 10x with the current amount of work and a subcription would save him alot of costs in the long run.