It’s not clear to me what auto does as far as picking a model.
Does it respect the list of models I enable in Cursor Settings > Models? Or will it pick e.g. gpt-4o even if I only enable gpt-5in that list?
And does it dispatch to different models based on how much capability it thinks my request needs, or is it entirely based on estimated cost, rate limits, etc.?
Hey, no, the auto mode selects models from the list of available premium models. It doesn’t depend on which models you have enabled. For more detailed information, see here:
Is there a list of these “premium models”? I assume it’s the likes of gpt-4o, sonnet-3.7, etc.? And that the most expensive models like opus-4.1 are excluded?
It seems to me that, when coding, it’s better to use a single model for consistency, rather than randomly jump between unknown models. Has that been others’ experience as well?
From that page:
“Once you find combinations that work well, like specific prompts paired with certain models, you can save them as Custom Modes.”
Unless I’m missing something, using Auto keeps us ignorant about everything described on that page.
It would be very useful if Auto would indicate which model is being used so that we can educate ourselves about the different strengths/weaknesses of various models.
You can think of it like this: You are the boss.
In Auto mode - You are handing over work to any junior employee because the task is trivial.
When you want work assigned to specific individuals, you can turn off auto mode and select a model.
When you want work assigned ONLY to a specific individual because the task is complex, then use custom mode and select a model and add instructions for it. Usually, your senior employees are chosen for these tasks.
Note: By senior employees, I mean Opus, O3, Sonette, thinking, etc. By junior employee, I mean cursor-small, deepseek, gpt-4o, etc.
This can help identify which model Auto is providing. I get Sonnet 4 most of the time. A Small Trick to Nudge the 'Auto' Model Toward Your Preference - #5 by MidnightOak