I Benchmarked Three Cheap Models on My Real Work

I ran my actual tasks, drafting, tagging, and summarizing, through three models that cost pennies.

On drafting, two were within a hair of the flagship for my purposes. On tagging, all three tied.

The flagship won only on the gnarly reasoning I hit maybe twice a week.

The leaderboard ranks a different job than mine. Benchmarks optimize for the hard question, not the common one.

I moved the common work down the price ladder and kept the top tier for the rare hard case.