If anyone has been avoiding auto, I’m on the ultra plan, and still run into the usage limit at half-a-month. I was afraid of using Auto because well, it was terrible for a while. But last month, I turned it on, and it’s doing a great job. Now I use auto primarily then switch to Opus or Sonnet or Haiku or Composer 1 if I want something that’s going to think a bit more. But auto now is fast, and generally as accurate as when I’m using Sonnet. Are other people finding this to be the case?
6 Likes
What I do is run everything in auto mode.
And I use gemini 3 thinking directly in a separate chat for hard thinking/planning.
And I use claude code in the powershell in cursor IDE for occasional hard thinking on the files. This way I can use the claude pro plan instead of API.
do we know what models auto routes to yet?
1 Like
we never know since Cursor hide it. in past, we can detect base on how much context auto have. but right now they change into percentage
Yeah, I’m using it in auto for a good time, sometimes I just change to use the max mode with gemini 3 for mapping and doc.
1 Like
Many people prefer the UI and features on cursor to Claude code, it is much better. Also nice to be able to mix it up when Claude gets stuck.
The only problem of auto is it’s a little bit slow.
Great news, auto isn’t free anymore so none of this matters anymore, spread it before anyone loses a lot of money on the auto model…
Auto picks cheaper models for quick tasks, I also feel like they are somehow subsidising auto over directly using premium models as I tend to get 5-10x usage out of auto vs using Claude LLMs directly.