I use the Auto model to save on request usage. Ideally, I’d like to stick with Claude Sonnet 4 all the time, but sometimes it’s just too slow. The Auto model can occasionally give decent responses, but other times it feels noticeably dumber.
Since the dashboard shows only “auto”, I couldn’t tell which model was used. I noticed that in Agent mode, the model sometimes tells you its name when asked — so I added this to my user rules:
**MANDATORY**: At the beginning of **every task or step**, you **MUST** clearly state the **model name, model size, model type, and its revision (updated date)**. This requirement is **non-negotiable** and must be followed **without exception**.
Yeah, it’s a bit verbose — but I had to be strict, because without this, the model sometimes skips mentioning its identity entirely.
I know models sometimes get confused about who they are (e.g., Claude 4 claiming it’s 3.7), but so far this works well for me. In my case, the model usually picks Claude 4 or GPT-4, and occasionally Claude 3.7. Interestingly, when GPT-4 is chosen, it tends to act a bit dumber in my experience. When that happens, I just open a new chat or pause the task until I get Claude 4 again.
Not sure if this is considered abuse. Just sharing in case it helps others.
