The basics

AI model

In short

An AI model is the trained core of an AI tool: a large set of learned numbers that takes an input, such as a question or a photo, and produces an output, such as an answer or a label. Apps add a chat window and other features around it.

Also called
Model, Machine learning model
Topic
The basics
Level
Beginner

A model is what comes out of training. A learning method is run over training data and fits the model's learnable parameters, the internal numbers that decide its outputs1. GPT-3 has 175 billion parameters2, and Meta's Llama 2 models range from 7 billion to 70 billion3.

Once trained, a model works without its training data. A 2025 High Court judgment on the image generator Stable Diffusion recorded that inference "does not require the use of any training data and the model itself does not store training data"4. The judge found the model weights were not themselves an infringing copy of the images used in training4.

A model is not the same as an AI system or app. The EU AI Act says models "do not constitute AI systems on their own" and need further parts, "such as for example a user interface", to become systems5. A general-purpose model can be "integrated into a variety of downstream systems or applications"1.

Models reach users in different ways. The EU AI Act lists libraries, APIs, direct download and physical copies, and notes models can be fine-tuned into new models5. Meta, for example, released its Llama 2 models for others to build on3.

Example

When you type a question into a chatbot, the app passes your text to a model and shows you its reply. The app is the system; the model is the engine inside it5.

Why it matters

News about AI often turns on models: which one a product uses, how big it is, and who can download it. The EU AI Act also sets separate rules for general-purpose models1.

Questions people ask

What is the difference between an AI model and an AI system?

A model is the trained component. A system wraps it with other parts, such as a user interface. The EU AI Act says AI models "do not constitute AI systems on their own"5.

Does an AI model contain its training data?

Not as a library of copies. In Getty Images v Stability AI (2025), the High Court found that Stable Diffusion did not store the copyright works it was trained on, and its weights were not an infringing copy4.

What does 70B mean in a model name?

It normally refers to parameters, the learned numbers inside the model: 70B means 70 billion. Meta's Llama 2, for example, came in sizes from 7 billion to 70 billion parameters3.

Related terms

Sources

  1. Regulation (EU) 2024/1689 (EU AI Act), Article 3, Definitions (artificialintelligenceact.eu mirror; EUR-Lex blocks automated access)
  2. Brown et al. (2020), Language Models are Few-Shot Learners, arXiv:2005.14165
  3. Touvron et al. (2023), Llama 2: Open Foundation and Fine-Tuned Chat Models, arXiv:2307.09288
  4. Getty Images (US) Inc & Ors v Stability AI Ltd [2025] EWHC 2863 (Ch), paras 7, 48, 556, 600
  5. Regulation (EU) 2024/1689 (EU AI Act), Recital 97 (artificialintelligenceact.eu mirror)

Back to the glossary