Why I stopped chasing every new AI model
A new flagship used to mean I had to try it that night. Then I noticed the upgrades stopped changing my output.
For two years I installed every new model the day it dropped. A new benchmark leader meant a new tab, a new API key, a new 'this changes everything' message to my group chat. None of it made my work noticeably better after the first week.
The truth I avoided: most of my tasks are writing, summarizing, and light coding. For those, the gap between a good model and a great one is smaller than the gap between a clear brief and a vague one. The model was rarely my bottleneck.
I now keep two: one cheap workhorse for bulk jobs and one strong model for the hard reasoning. When a new release lands, I run one real task through it before I care. Most times it is fine and I move on.
Chasing models felt like progress because it was easy. Actually improving meant writing better prompts and judging outputs, which is harder and pays off more.