I like that you addressed price for the first one.  This is definitely a consideration for me.

Where is the link you mentioned?

Is there a LLM for just math or stats?

For local models such as Llama how can we the effectiveness as a whole for smaller vs larger models?

Mixtral is the best for local use on regular/high performance computer

on device and makes it free and is good enough for most tasks. 
stuff like gemini nano is even multi modal on some devices.

Great

In the description

🤨