Follow AI news for a week and you'll see a stream of new model names. It looks chaotic, but most names follow a few simple patterns.
Most labs give their models a family name and a number. The family is the product line and the number is the generation, a bit like phones, where each new model gets a higher number. In-between numbers, like a 4.5 after a 4, usually mean a big update within the same generation.
Labs often release a big model first and smaller, cheaper versions later, or launch a whole range on the same day. So a single launch can add several new names at once.
Reading the names
The word after the number usually tells you the size or tier, like the different versions of the same car, from basic to top of the range. Pro, Ultra and Max usually point to a bigger, more capable version that's slower and costs more to use. Flash, Lite, mini and nano point the other way: smaller, faster and cheaper. Some nano models are small enough to run on a phone.
Turbo usually means a faster, cheaper version of a model that already exists. Thinking means the model works through a problem step by step before it answers, which is slower but better for maths, code and tricky questions. A date at the end tells you when that version was made and usually marks a Snapshot: A dated, frozen version of a model, so developers' apps keep behaving the same even when the main model is updated.: a frozen copy that developers can rely on not to change.
On models that are free to download, you'll often see Instruct, the version trained to chat and follow instructions, and Base, the raw version that developers build on. If you just want to try one, pick Instruct.
Fast, cheap or clever
Why not use the biggest model for everything? Bigger models are slower and cost more to run, and most everyday jobs don't need them. It's like hiring: you wouldn't pay a senior lawyer to sort the post.
So labs offer a range, and developers pick the cheapest model that does the job well enough. A small model might sort thousands of support emails while a big one tackles the hard questions. Some apps switch between models behind the scenes, so you may use several without knowing it.
Specialist models
Not every model is a chatbot. Image and video generators make pictures and clips from a description. Voice models turn text into speech or speech into text, and there are models for music, coding and controlling robots.
Some useful models are ones you never talk to. Embedding model: A model that turns text or images into lists of numbers so software can find similar things. It powers search and recommendations; you don't chat with it. turn text into lists of numbers so software can find things with similar meanings. That's how the search box on a help site can find the right article even when you typed different words. They quietly power search and recommendations.