Fundamentals

AI Model

An AI model learns patterns from large amounts of data and then predicts or generates output from new input. It is the engine inside a product such as ChatGPT.

The engine, not the product

An AI model learns patterns from large amounts of data and then predicts, classifies, decides, or generates text and images from new input. In a conversational assistant, it reads your question and produces an answer by predicting what should come next.

From the outside, the model is the engine running inside a service. ChatGPT is the product; GPT is the model inside it. Claude uses similar names for the product and the model family, and Gemini uses the same name for the app and the model, which is why the distinction is easy to lose.

Service, model, and tools

  • Service: the product — chat interface, history, files, search, and pricing
  • Model: the core that interprets input and generates the answer or image
  • Tools: extras the model can call, such as search, code execution, image generation, or external apps

Calling the same model through an API does not necessarily reproduce what the product gives you. The service’s own instructions, search, memory, connected tools, and safety settings all shape the answer.

Kinds of models

There are large language models for text, image generation models, speech recognition models that turn audio into text, and video generation models. Models that handle text together with images, audio, or video are called multimodal.

Large models that serve as a base for many uses are called foundation models; those tuned to work through hard problems step by step are called reasoning models. These are not separate boxes — one model can have several of these properties.

Choosing one

  • Does it handle the kind of input you have — text, images, code?
  • Do you need speed, or depth of reasoning?
  • How much source material can it read at once?
  • Can it use the tools you need, such as search or file creation?
  • Do the pricing, usage limits, and data handling fit your use?

A higher number in the name, a newer release, or a better benchmark score does not make a model right for every job. Matching the model to the weight of the task — something fast and light for small edits, something capable for serious research or design — keeps both cost and time down.

Related terms

Sources and review information

Last reviewed July 20, 2026

Back to the AI Glossary

Search this site