A model is what training produces: a large set of learned numbers arranged in a particular structure. Products such as chat apps, image generators and voice tools are built around one or more models, and many let you choose between faster, cheaper models and larger, more capable ones.
How it works
A model has two parts. The architecture is the blueprint, such as a transformer for language or a diffusion model for images, describing how data flows through layers of calculations. The parameters, often called weights, are the numbers inside that blueprint, set during training on data. Small models may have millions of parameters; frontier language models have hundreds of billions or more. A trained model is saved as files containing those weights, which can be gigabytes to hundreds of gigabytes in size. Using it, called inference, means feeding in an input, text, an image or audio, converted into numbers, and computing an output. Companies release models in versions and sizes, for example a large flagship, a mid-sized option and a small fast one, trading quality against speed and cost. Some models are closed, available only through an app or API, while open-weight models publish their weights so anyone can run or adapt them. A model can be further adapted with fine-tuning, or steered at use time with prompts and system instructions.
Why it matters
The model you choose sets the quality, speed, cost and privacy of your results. A small model may be ideal for tagging thousands of product images cheaply, while a larger reasoning model suits a complex strategy document. Knowing that apps are wrappers around models helps you compare tools fairly, understand why the same app changes behaviour after an update, and decide when an open-weight model you can run yourself makes sense.
Where you will see it
- Picking a fast, low-cost model to write first drafts of social captions in bulk.
- Switching to a larger reasoning model for analysing a complex contract.
- Running an open-weight model on company servers to keep sensitive data in-house.
- Using a specialised speech model for transcription and a separate model for summaries.
Limits and common misconceptions
- Bigger is not always better: larger models cost more and run slower, and a smaller one may be good enough.
- A model is not the same as the app; two apps using the same model can give different results.
- Model behaviour changes between versions, so a prompt that worked last month may need adjusting.
How it differs from related terms
- Model vs Algorithm An algorithm is the step-by-step procedure, for example the training method; the model is the trained result that procedure produces.
- Model vs Parameters Parameters are the learned numbers inside a model; the model is the whole system, architecture plus parameters.
Frequently asked questions
What is an AI model in simple terms?
It is the trained part of an AI system that does the actual work. Think of it as a very large formula whose numbers were tuned by learning from data. You give it an input, such as a question or photo, and it calculates an output, such as an answer or caption.
What is the difference between an AI model and an AI tool?
The model is the engine; the tool is the vehicle built around it. A tool adds an interface, file handling, memory, integrations, safety filters and other features. One model can power many different tools, and one tool may switch between several models for different tasks.
Why do AI apps offer different models?
Models trade off speed, cost and capability. Smaller models are quick and cheap for routine tasks. Larger or reasoning models are slower and more expensive but handle complex, multi-step problems better. Offering a choice lets users match the model to the job and their budget or usage limits.
Related terms
- Machine learning A branch of AI in which computers learn patterns from data to make predictions or decisions, instead of following rules a programmer wrote by hand.
- MCP (Model Context Protocol) An open standard, introduced by Anthropic in November 2024, that defines how AI applications connect to external tools, data sources and services.
- Multimodal AI AI that can take in or produce more than one type of data, such as text, images, audio and video, and connect information across them.
- Natural language processing (NLP) The field of computer science and AI concerned with enabling computers to process, understand and generate human language, written or spoken.
- Neural network A machine learning model made of layers of simple connected units, artificial neurons, whose connection strengths are adjusted during training to learn patterns from data.