Does Auto include only what you have checked in Preferences=>Cursor Settings=>Models or is it irrelevant to it?
Maybe @deanrie knows?
Does Auto include only what you have checked in Preferences=>Cursor Settings=>Models or is it irrelevant to it?
Maybe @deanrie knows?
Hey, no, it selects from the premium models, except for o3. More detailed info here:
Thanks, @deanrie! This link mentions (in “the pool above”) only 3 premium models (see following cnfirmation).
Why not gemini-2.5-flash, gpt-4o, grok-3-beta, grok-3-mini, o3-mini, o4-mini, etc.?
Also, does “auto” use thinking only when it wants to?
It would be nice to add “auto” icon to the models list in docs so we can see which models are supported by auto.
Auto mode selects models only from the premium segment and does not use models with thinking capabilities.
Can you please confirm that GPT 4.1 is among the auto models?
GPT 4.1 is on the “pool above” listed in the docs, but one of Cursor ambassadors on Discord claims GPT 4.1 is not among the auto models.
For the context, I consistently get auto model to identify itself as GPT 4.1.
I agree it should be mentioned.
The issue, @deanrie, is that the docs don’t have a “premium segment” section. So are we talking only about the Selection tree?
As far as I know, and according to the documentation, Cursor’s Auto mode uses a model pool that includes gpt-4.1. The only exception is the o3 model. gpt-4.1 is classified as a “reliable” model, along with claude-4-sonnet and gemini-2.5-pro.
The documentation isn’t clear, @deanrie, since it just mentions “the pool above” which I guess really is “the Selection tree”. Feels like Treasure hunting.
Why are all these considered unreliable?
I’ll discuss this with the team to see if we can clarify more specifically which models are included in Auto mode.
Thanks, @deanrie, please document which and why.
I think showing which model is currently using in Auto mode and show why this model is selected inside IDE is the best.
Why still GPT-4.1 rather than its successor GPT-5 that came out earlier this month? Unless it did switch to GPT-5 without updating the documentation?
The whole “selection tree” was removed altogether from the docs.
”Auto” was also rephrased to include “the premium model best fit for the immediate task”.
Does it mean it returned to using not just the 3 aforementioned “reliable” premium models, @deanrie?
The OP has made a GREAT feature request … this is how auto should really work !!!
Only the models I prefer should be what Auto consists of, not what the Cursor overlords think i shoulld be using
Totalitarian model selection is not cool anymore