There is not one model

There are many models from many makers, they change constantly, and the differences are big enough to matter.

Part of the What is this thing? track on lAItest.

There is no such thing as "the AI". There are dozens of models from a dozen makers, and the leaders change every few weeks.

Here is the board, as of right now.

Four things you can compare without any expertise

How much text fits at once. What it costs per million tokens in, and per million out. When it was released. Whether anyone can download it and run it themselves, or only rent it through the maker. Every one of those is published by the makers. Everything else — how clever it is, how well it writes — is contested and measured badly.

The spread is not small

Try it

Sort by price, then by window, then by date. Notice that no single model wins every column. This step is an interactive widget; open the lesson to use it.

A common misconception

Commonly believed: One model is the best one, and the sensible thing is to use that.

Actually: The columns disagree with each other. The largest window is rarely the cheapest. A newer flagship has shipped with half the window of the model it replaced. A newer, cheaper model from the same maker can be worse at long, repetitive work than the older one it appears to supersede. Choosing is a fit question — what you are doing, how much text, how much money — not a ranking question.

In one sentence

"The AI" is a crowd, it reshuffles constantly, and the durable skill is knowing which four numbers to go and look up.