Transparency on "auto" model selection

I like the “auto” feature but I’d like to know what model was selected for a given request. In particular, if I didn’t like the result I’d like to know if it could be due to the model choice, so I could try again with an explicitly differnet model.

9 Likes

agreed. each model has its strengths and weaknesses. knowing which one did (or didn’t do) the task would help tons with efficiency and productivity.

I agree. I leave mine on auto most of the time, and it’d be really nice to see which model was used for which response. An information (i) would be really awesome with useful information about which model, inference times, tokens, context, etc.

On the same note, having a way to see how much context is used/left would also be really nice. I often can tell when it appears to have reset context, but it’d be nice to know for sure so I can realign the agent with things it forgets with context drops.

1 Like

I registered on the forum EXCLUSIVELY to ask for this. I am alright with the auto select, but need transparency in knowing what model is chosen!

1 Like

I did exactly too!

:100:
I’ve never used “auto” for exactly this reason.

Auto-select

Enabling Auto-select configures Cursor to select the premium model best fit for the immediate task and with the highest reliability based on current demand. This feature can detect degraded output performance and automatically switch models to resolve it.

I’m fine with it choosing the ‘best fit’ model, but without knowing what was used there’s nothing to learn from and no way to judge. :man_shrugging:

The ‘current demand’ part is where I get a bit cynical that it’s more about reducing cost vs. best results for the task.

With transparency this could be beneficial allowing users to opt-in to using non-premium models sufficient for the task if it reduces requests used.

Also, the docs mention ‘premium model’, but they are no longer described that way.
Does Premium =1, >=1, or <2 requests?

.
(P.S. A ‘trick’ I just thought of but haven’t tried is including instructions to write the model name in a code comment.)

AGREED, main reason i never choose auto

yess!