Model types: chat, thinking, pictures, embeddings, speech

Not every oven bakes the same thing. Some write, some think, some paint, and some turn text into numbers. The ModelsModels page sorts them by type; here's what each one is for.

Chat ovens

Chat models read text and write text back: emails, answers, summaries, posts. Many can also look at pictures you attach, like a photo of a receipt. These are the ovens your bakers use most.

Thinking ovens

Reasoning models are chat ovens that think quietly before they answer. They're better at puzzles, plans and long sets of rules, and slower and pricier. The lesson on thinking effort shows a real example.

Picture ovens

Image models paint a picture from a description. They're priced per picture, by size and quality. This one was painted by a picture oven from one sentence, "a cozy illustrated kitchen with a grandmother baking a pie, warm light":

Embedding ovens

Embedding models don't write words at all. They turn a piece of text into a long list of numbers, an embedding, that captures its meaning. "I love cake" and "Cake is my favorite" get very similar numbers; "The car is red" gets quite different ones. That's how the Pantry finds the right page in your documents. Try it yourself on the Test Kitchen's Embeddings tabEmbeddings tab.

Speech ovens

Speech models read text out loud in a chosen voice, and transcription models do the opposite: they turn a recording into written text. Both are priced by the minute or by the amount of text.

Granny's rule of thumb: pick the type first (words, thinking, pictures, meaning or voice), then the cheapest oven of that type that does the job well.