17 października 2026
5 min czytania
Opinion
I Stopped Chasing the Flagship Model and My Bills Went Down
For most of what I do, the difference between a frontier model and a small one is invisible in the output and very visib
I clicked around for ten minutes and this is what stuck.
For most of what I do, the difference between a frontier model and a small one is invisible in the output and very visible in the invoice.
What I would do differently: Claude Haiku 5.5 and GPT-6 Luna handle my daily drafting and sorting at a fraction of the cost of the top tier, and I cannot tell the difference in a tweet.
The bit I actually liked: I now reserve the expensive models for the two or three tasks a week where reasoning actually breaks. Everything else runs on the cheap tier.
I will probably forget it exists in a month.