AI-modell

in English: AI model

basicwidespread

An AI model is a program trained on countless examples that responds, predicts, recognizes, or creates new content based on learned patterns.

In plain terms

The AI model is the "engine" of artificial intelligence. When you chat with a bot, generate an image, or your phone recognizes your face, a model works in the background. The application is merely the body: buttons, windows, and input fields.

It is like a chef who has tasted thousands of dishes and grasped the ratios. That is why it can prepare a dish it has never seen before. The model does not primarily memorize examples but the patterns distilled from them, although it can sometimes recall frequently seen texts almost word for word.

This differs from traditional programming. There, the programmer writes down the rules step by step, whereas in the model, the rules emerge from the data during training. This is why it handles tasks that are difficult to codify, such as text understanding or image recognition.

Many types of models exist. Some write text, some generate images, audio, or video, and some do only one thing: filter spam or estimate the risk of a loan.

How it works

Data collection: developers gather huge amounts of examples, such as texts, images, audio recordings, or tabular data.
Training: the model repeatedly attempts to solve the task using examples and, based on its errors, makes small adjustments to its internal settings, the so-called parameters. In the largest models, there can be hundreds of billions, or even more than a trillion, of these.
Freezing: at the end of training, the parameters are frozen. The resulting model is essentially a huge set of numbers carrying the learned patterns.
Usage: when you provide a question or image, the model calculates the output from the input using its parameters. For example, a language model calculates the probability of each possible next text segment and selects from these, so the answer is assembled piece by piece.
Integration: the model is embedded into applications, services, or agents, where it receives a search engine, memory, and other tools, becoming a ready-made product.

An everyday example

You ask the chatbot to draft a polite complaint letter to your service provider, and within half a minute you receive a ready text. The letter was not written by the application itself but by the language model running behind it. The same model may be present in products of other companies, under a completely different name and interface.

Who is affected

everyoneFrom search engines to translators to phone cameras, you use models daily, even if you are unaware of it.

employeeMore workplace tools are backed by models, and the quality of your work also depends on how well you know their limitations. → Work & HR

entrepreneur, executiveYou must decide which model your company's processes are built on, and this determines costs, data security, and quality alike.

developerModel selection, fine-tuning, and integration have become everyday parts of software development. → Code

A common misconception

MythAn AI model is a massive database from which it retrieves the ready-made answer.

In realityThe model does not store the training data verbatim nor search through it. It generates answers anew each time based on learned patterns, which is why it can convincingly err. It may sometimes reproduce certain frequently repeated text segments almost word for word.

Limits and risks

The model can state falsehoods with confidence, so always verify important facts from other sources.
Its knowledge ends when its training was completed; it only knows about later events if it is given a search tool or a recent document.
It absorbs biases from training data, so its decisions may be unfair toward certain groups.
Often even developers cannot say exactly why a specific answer was given, which makes error detection and clarifying responsibility difficult.

Similar, but not the same

AI-rendszer – The model is only the core, while the AI system is the full product built around it, including the interface, data, and security rules.
Algoritmus – The algorithm is the recipe according to which training occurs, while the model is the result of training, the finished 'knowledge'.
AI-agent – An agent is built on a model but also receives tools, executing tasks independently in multiple steps, whereas a model alone only provides output for an input.

Related concepts

Broader

Artificial intelligence, Machine learning, AI-rendszer

Narrower

Large language model, Foundation model, General-purpose AI model, Image generation model, multimodal model

On the edge

Training data, Parameter, Neural network, Fine-tuning, Inference

A brief history

One of the first examples of learning models was the perceptron described by Frank Rosenblatt in 1958. The foundation of today's large language models is the transformer architecture introduced in 2017, and the general public became familiar with them when ChatGPT launched in November 2022.

Where things stand

By the end of 2026, models are standard components of everyday software, from phones to office programs. In the EU, the AI Act sets out specific rules for providers of general-purpose AI models; these obligations have applied since August 2025.

Status: 2026 November

In the news

Appears in 86 Agent Age stories; the curve shows weekly occurrence in the news magazine.

Visa test: AI shopping agents failed 42.8% of runsNov 5, 2026 · Money
AI models may predict sovereign defaults years ahead, and credit insurance could benefitNov 5, 2026 · Money
Multimodal delirium risk model validated externally and deployable without site-specific retrainingNov 5, 2026 · Health
GitHub Copilot Fusion picks models automatically; local preview runs only partly on-deviceNov 2, 2026 · Code
Chinese robots and open AI models could reshape Hungarian jobs tooOct 30, 2026 · Ipar
Bank of America and Capital One use smaller AI models to cut costsOct 28, 2026 · Money
Stanford authors sue Anthropic over use of copyrighted books to train AIOct 28, 2026 · Jog
AI political ad spending in US midterms tops $137 million as regulation lagsOct 26, 2026 · Marketing

Also known as

AI-modell · MI-modell · AI modell · AI model · machine learning model · AI model · machine learning model · ML model

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.