Kinds of model
What a model is built to do.
General-purpose models
General-purpose models are the large language models built to do a bit of everything: answer questions, write and edit, summarise, translate, reason through a problem and write some code. A language model is a neural network trained to predict the next token, a word or a piece of a word, across huge amounts of text
Reasoning models
Reasoning models are language models trained to work through a problem in a long chain of intermediate steps before they give a final answer. The extra thinking tokens cost time and compute each time the model is used, but they lift results on mathematics, science and coding
Code models
Code models are language models trained or tuned to read and write software. They complete a line as a programmer types, turn a plain-English request into a function, explain unfamiliar code, write tests and, in their newest form, act as agents that edit a whole repository and run its tests
Multimodal models
Multimodal models take in more than text. Most read images, such as photos, charts, handwriting and scanned documents; many also listen to speech, watch video or produce images and audio themselves
Small and on-device
Small language models trade breadth for speed and cost. With from a few hundred million to around ten billion parameters, they can run on a laptop, a phone or a single modest graphics card, keep data on the device and answer without a network connection
Domain models
Domain models are language models built or tuned for one field, such as medicine, law, finance, science or robotics. Some are trained from scratch on a field's own text; more often a general model is fine-tuned on specialist data and tested against the field's own examinations
Regional and language models
Regional and language models are built for a particular language, country or region. Most large models learn mainly from English web text, so they can be weaker in other languages, use more tokens for them and miss local culture and law
Encoders and embeddings
Encoders and embedding models turn text into meaning rather than writing new text. An encoder reads a whole passage at once and produces a vector, a list of numbers, for each word or for the passage, placing similar meanings close together
Research milestones
Research milestones are models that changed how the field works, whether or not anyone still uses them. Some introduced a method, such as pre-training followed by fine-tuning; some showed what scale could do; others corrected earlier thinking, such as the 2022 finding that most large models had been trained on too little data
Assistants and chatbots
Assistants and chatbots are the products people actually talk to: an app or website wrapped around one or more language models, with memory, search, file handling, voice and safety filters added. One assistant often switches models behind the scenes, and its name can outlive several model generations