Why does Cursor still have an 'Auto' model?

Discussions
,
1
  • We have a frontier casino model - Premium.
  • We have Composer 2.5, which outshines most models in the Auto roster, and often appears in Auto itself.

So why do we need a full-roster gacha mode at all?

I didn’t understand the point of this mode from the very beginning, and after Composer 2 it became doubly pointless :thinking:

4

Fair point that Composer 2.5 is outstanding — but I’d reframe it: Auto isn’t about finding the single best model, it’s about hedging against context, no? Some tasks favor Haiku’s speed, others need Sonnet’s depth, no? Even with a clear winner, selection has value. That said, if Composer 2.5 is already appearing in Auto frequently, maybe the system just needs smarter weighting, not removal? hmmm…

7

Hi @Artemonim Good question. There’s still a few cases where Auto is really helpful and I’ll give you a few scenarios:

  • A model provider is under heavy load, and it makes sense to route requests over to another model, even if that model is more expensive or not used as much (this happens seamlessly with Auto vs. getting a “Too many requests/High load error” in the client.
  • A user has changed regions / is traveling, and the model that they typically use isn’t showing up in the model picker.
  • A new user downloads Cursor → has never used the product before, and is overwhelmed by the variety of options for which model to choose to build their first project.

In all of these cases (and more), Auto is a great option! But totally understand if it’s not for you and you prefer other options!

8

I don’t visit the forum very often, but I keep seeing newcomers complain about the Agent performing poorly, when they’re using Auto mode — and there’s a good chance that switching to Composer or Premium Router would solve the problem :thinking:

Personally, I don’t really see any meaningful middle ground between Composer 2.5 and GPT-5.4/5.5 right now. You could throw Sonnet into the mix, but why? Opus is somewhere in that range, but it’s clearly more expensive and seems better suited for specific use cases. Gemini sucks, and Grok just does it even better…

9

I am missing transparency: Why does the auto mode not reveal the model being used at a given point in time?

10

^ This is the primary reason I don’t use Auto at all for actual code writing tasks (I’ll sometimes use it to generate code documentation since that’s not executable). I am fine with the notion of “pick something for me” but I would like to know what was chosen at some point.

11

Lots of debate and lengthy discussions on this point over the years.

It’s a deliberate Cursor design choice not to reveal the model Auto selects.

12

Why not be transparent in why not being transparent in disclosing the chosen model?