I’m a bit concerned about using Auto. If I select Composer 1.5 or Composer 1 it seems to complete tasks fairly quickly but when I use Auto it often seems to use a verbose model that does a lot of thinking. This can go on for quite some while and it makes me wonder if I’m burning through tokens more quickly using Auto than I would manually using 1.5 or 1 or something else (e.g. Sonnet).
I’m concerned that from a speed (it takes longer to do all the thinking) and from a cost perspective Auto may not be working for me.
I’m not doing anything hugely complex usually. It would be nice to be able to select which models can be used in Auto mode.
I agree with you. Right now, Auto and Composer share the same usage pool, so usage is combined between them.
You’re also right that Auto can feel more verbose. In some cases, it may burn more tokens because it tends to “think” more before responding. However, Auto output tokens are priced at around $6 per 1M tokens, which is generally cheaper than most advanced or smart models. So even if token usage is higher, the per-token cost is lower.
Regarding speed, I’ve also noticed that Auto sometimes behaves like it’s using GPT-class models, which tend to reason more deeply and produce more detailed outputs. That can make it feel slower compared to manually selecting Composer 1 or 1.5. So your concern about speed and cost is valid — especially if your tasks are usually not very complex.
From a product and pricing perspective, Auto is designed to abstract model selection and operate under a fixed token cost (e.g., $6 per 1M output tokens).
Because of that fixed pricing model, giving users control over which models Auto can use could make cost predictability and internal optimization more complex. Auto is likely optimized for balancing performance and cost behind the scenes rather than exposing that layer directly.
It would definitely be useful to have more control in the future, especially for users who prefer optimizing for speed or specific models.
@Naufaldi_Rafif covered this well! The key point is that Auto’s output rate ($6/1M tokens) is actually cheaper than both Composer 1 ($10/1M) and Composer 1.5 ($17.5/1M). So even though Auto may produce more verbose responses with thinking models, the per-token cost is lower - meaning you’re likely not burning through your usage as fast as it might feel.
That said, your concern about speed is totally valid. If you’re doing straightforward tasks and want snappier responses, manually selecting a faster model like Composer 1.5 is a solid approach. Quick tip: you can use Cmd/Ctrl + / to quickly switch between models without leaving the chat.