This appears to be written by someone who isn't a Cursor user.
That product does not silently switch models under the hood. The user is able to see and choose the model for any prompt before it is submitted.
What Cursor does do under the hood is to dispatch tools to use a smaller model for subtasks, currently Composer 2.5. But no one cares about that. It's for janitorial and brief investigative work, nothing more.
Regardless of who owns Cursor, it will be a harness for working with multiple agents and models and not the property of one model maker. That's exactly what makes it different.
isn't that totally unprovable given what the user sees, given that its silent? if they wanted to do this it would be super easy to just fully replace the larger model with its MTP counterpart for a few seconds, and if done correctly no one would notice.
Those people aren't the customer. This is a tool for professional devs who know what to expect from each model.
They got their customers by being what they are. It doesn't make any sense that they would throw that all away to pointlessly mislead people. Their entire business model is selling tokens to access these models.
Hm. This article is riddled with assumptions that seem to me only to hold if you pretend that neither openrouter and open source tooling exist?
I don't use Cursor and I was never going to, so what, significantly, am I missing, given that the article's suggestion is spending two to four weeks baking one's own abstraction layer?
This appears to be written by someone who isn't a Cursor user.
That product does not silently switch models under the hood. The user is able to see and choose the model for any prompt before it is submitted.
What Cursor does do under the hood is to dispatch tools to use a smaller model for subtasks, currently Composer 2.5. But no one cares about that. It's for janitorial and brief investigative work, nothing more.
Regardless of who owns Cursor, it will be a harness for working with multiple agents and models and not the property of one model maker. That's exactly what makes it different.
> not silently switch models under the hood
isn't that totally unprovable given what the user sees, given that its silent? if they wanted to do this it would be super easy to just fully replace the larger model with its MTP counterpart for a few seconds, and if done correctly no one would notice.
No, because the purpose of the product is to act as an agent harness and to dispatch tasks to the correct agent at the correct time.
Messing that up would make it an entirely different kind of product, and not one that people want.
"the people" cant tell if they use a smaller/cheaper LLM when cursor is writing a simple file
Those people aren't the customer. This is a tool for professional devs who know what to expect from each model.
They got their customers by being what they are. It doesn't make any sense that they would throw that all away to pointlessly mislead people. Their entire business model is selling tokens to access these models.
Hm. This article is riddled with assumptions that seem to me only to hold if you pretend that neither openrouter and open source tooling exist?
I don't use Cursor and I was never going to, so what, significantly, am I missing, given that the article's suggestion is spending two to four weeks baking one's own abstraction layer?
I love how there's a section called "The SuperML Take" when the whole thing is obviously written start to finish by Claude or GPT.
I'll save you a click: it'd be bad if you can no longer choose what underlying model to use in Cursor after the SpaceX acquisition.
Not worth the time to read it.