What is the ratio/overhead of queries using Claude in Cursor vs using Claude CLI for the same $200? I would much prefer Cursor since I can switch models when the chosen models get stuck, or if I want a second opinion, but what is the cost of this flexibility? If the number of queries is comparable, then Cursor is the clear winner.
Is Cursor more efficient than Claude CLI at tokens for the same task? I would expect so since Cursor has an incentive to be more efficient.
I’ve had far better results with Claude code.
I notice the same but assumed it was that claude cli was just using a newer model. A single week makes a big difference.
Usage is not comparable at all, Claude Code is far more requests per month. Their rate limits reset every 5 hours, you have fixed use on Cursor $200 plan ($200 in token usage total for the month) - so from a cost/request or how many requests you can submit, Claude Code is by far the better value.
However, you said yourself you prefer working with Cursor, so you’ll have to decide if the increased cost if you need more than $200 usage per month is worth it in Cursor. While Claude Code might provide many more requests, it doesn’t mean you’ll be happy in the terminal, or Cursor interface might make the $200 Cursor plan preferable to you. It’s really an individual choice, but on pure “how many requests can I do in aggregate over a month for the $200 spend” there’s not even a question. Monitoring my use on $200 claude code plan last month, I easily spent well over $2,500 in equivalent API usage - not possible on Cursor $200 plan.
You may not know that Claude Code can be run in a terminal integrated within the IDE, often that’s what I do (or in another terminal but linked to IDE) so I can see the diffs in Cursor and when the occasion arises, I use my 500 requests/mo on cursor to get a different models perspective.
I should also point out that the above isn’t a knock on Cursor. They’re an independent company, they have investors who expect the company to grow (and eventually, make money). It’s not realistic to expect Cursor to offer a subscription for $200 that costs them $2k+ in credits to service, there’s no world where that business model is sustainable.
So clearly, they’ll have to keep the other tools and the “experience” of their chat window over the “terminal experience” such that customers are willing to pay for the higher costs. Its just not realistic for them to compete and win against claude code on pure usage limit assessment.
Claude CLI will not give you access to the other models. Claude isn’t an ultimate LLM. Also, Grok-4 and GPT-5 will soon be released
@Artemonim - Sure, there might be other frontiers that are better at certain tasks/benchmarks, and new ones coming. I don’t think OP or I am saying claude code models are 100% better or head and shoulders above all others.
What is being said, is that the max subscriptions offered by Anthropic are a value that can’t be matched based on usage that the fixed cost sub affords. You might prefer others, but many devs prefer the anthropic models right now as daily driver, the crazy usage the max plans allow just further sweetens that.
I just realized this a couple of days ago. I was totally convinced that Anthropic models are superior and I only ever need them (I was using Claude Code on Max). However, a problem that can’t be resolved by Sonnet/Opus was easily resolved by Cursor using o3 (I hit Sonnet 4 limit so I had to choose other models) in one go! The ability to choose another model is a huge advantage Cursor provides.