open-source AI

basicwidespread

Artificial intelligence whose model, and ideally the details of its training, is published by its creator for free use, allowing anyone to download, examine, modify, and run it on their own machine.

In plain terms

Closed AI is like a restaurant: you get the finished dish, but you cannot see into the kitchen and do not receive the recipe. In fully open AI, the recipe is yours: you can cook it at home, modify it, and share it further.

In practice, this means you do not access the model only through a company's website or app. You can download it and run it on your own computer or your company's server, so your data does not leave your premises.

Many researchers, small companies, and state institutions build on this because they do not depend on a single provider's prices and decisions. Moreover, independent experts can examine the open model, which helps uncover errors and biases.

The word 'open' covers many things. In some cases, only the final model is available, and you know nothing about the training data; elsewhere, the license restricts commercial use. It is always worth checking exactly what you get.

How it works

A company, university, or community trains a model, which requires vast computing power and money.
The developer publishes the finished model's "weights" (the billions of numbers formed during training, or hundreds of billions for large models) in a public repository, ideally together with the code to run it and a description of the training.
Publication comes with a license specifying how you can use the model: for research only, for business purposes with restrictions, or freely for anything. In the strict sense, only the latter counts as open.
Anyone can download it, run it on their own machine, or further train it with their own data, for example on Hungarian legal texts or a company's internal documents.
The community creates smaller, faster, or specialized versions of it and often shares these as well. Thus, an entire "family" can emerge from one model in a short time.

An everyday example

An accounting firm would not want client data to end up on foreign servers. Therefore, it downloads an open model and runs it on its own office computer, so nothing goes out to the internet while summarizing invoices or drafting letters.

Who is affected

entrepreneur, executiveWith an open model, you can keep sensitive data in-house and avoid dependence on a single provider's pricing, but in return, operation is your responsibility.

developerYou can freely experiment, further train the model, and build your own application or agent on it without a subscription. → Code

parent, student, teacherYou can try it for free, even without internet, and get a closer look at how AI works. → Education & Family

lawyer, citizenOpen models can be reviewed by independent experts. The EU AI Act partially exempts models with free and open-source licenses from certain obligations, but this exemption does not apply to models posing systemic risks. → Law

A common misconception

MythIf an AI model is available for free download, it is open-source and can be used for anything.

In realityMany 'open' models only have their weights available, not their training data or training details, and the license often restricts usage. Downloadability and true openness are two separate things.

Limits and risks

Running larger models requires powerful, expensive hardware; only smaller, weaker variants work well on an average laptop.
There is usually no customer support or warranty (unless you pay a provider for it): setup, updates, and security are your responsibility.
Security safeguards built into a downloaded model can be removed, making it easier to misuse.
Read the license before use, as the 'open' label often hides business or other restrictions.

Similar, but not the same

open-weight model – Only the trained model can be downloaded here; the training data and the training method may remain hidden, so in the strict sense this is not full openness.
free AI service – You can use a free chatbot, but you cannot download it, inspect its internals, or modify it.
open-source software – With a traditional program, the operation is readable from the source code. With AI, weights and training data are needed alongside the code, and even with these, it is not fully transparent why the model gives a specific answer.

Related concepts

Broader

artificial intelligence, open-source software

Narrower

–

On the edge

open-weight model, helyben futtatott AI, small language model, large language model, fine-tuning, closed model, digital sovereignty, licenc

A brief history

The term 'open source' was coined in 1998 in the software world. The first official definition tailored to AI, the Open Source AI Definition 1.0, was published by the Open Source Initiative in October 2024, after a heated debate arose over which models could truly be called open.

Where things stand

At the end of 2026, open models are widely used: American, Chinese and European developers keep releasing downloadable models, and in many everyday tasks these come close to closed systems. However, debate continues over what truly counts as open and how regulation should treat them.

Status: 2026 November

In the news

Appears in 20 Agent Age stories; the curve shows weekly occurrence in the news magazine.

DeepSeek's agent platform reaches 1.54 million daily users as Chinese open models spreadNov 2, 2026 · Agent Age
HSBC to open AI institute in Hong Kong following Singapore launchNov 2, 2026 · Money
Chinese robots and open AI models could reshape Hungarian jobs tooOct 30, 2026 · Ipar
Mistral Large 4 is the best open model outside China, with weights released publiclyOct 29, 2026 · Agent Age
European financial firms move to open AI models run on their own infrastructureOct 22, 2026 · Money
Mistral unveils 1-trillion-parameter Large 4, plans open-weights releaseOct 6, 2026 · Agent Age
Together Link runs open models inside Claude Code; Together AI says coding costs could fall by over 50%Oct 5, 2026 · Code
Meta launches open-source AI hardware project for makers under Apache 2.0 licenceOct 3, 2026 · Code

Also known as

open-source AI · open-source artificial intelligence · open AI · open model · open-source AI · open source AI

Sources

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