Models · 28 July 2026 · 4 min read
There Still Isn't One “Best” AI Model
The leaderboard changes every few months and none of it survives contact with what you actually do all day.
Ask which model is best and you will get a confident answer that is out of date within a quarter. The frontier labs leapfrog each other constantly, and the gaps are narrow enough that the winner depends entirely on the question.
Different models are good at different things
- Careful long-form writing and code review: one family tends to lead.
- Fast reasoning, tool use and general utility: another.
- Very long documents, multimodal input and cheap bulk work: another again.
- Open models: increasingly good enough, and dramatically cheaper for routine tasks.
So why does everyone pick one?
Because subscriptions force the choice. You pay one company twenty pounds a month and that decision quietly becomes the model you use for everything, including the things it is worst at. It is a billing artefact, not a judgement about quality.
The better default
Route the question to whichever model handles that kind of question best, and stop thinking about it. Most people do not want to become model selection experts. They want the right answer.