AI glossary · Models
Open-source AI
In one sentence
Open-source AI is AI whose model is released so anyone can use, study, modify and share it. In practice, many so-called open models publish only their weights, not the training data or all of the training code.
Also known as: open model · open-weight model · open weights · open-source LLM
What is open-source AI?
In software, “open source” has a precise meaning: you can use, study, modify and redistribute the program. Applying that to AI is trickier, because a model isn’t just code. It has three parts: the code that trains and runs it, the data it learned from, and the weights, the parameters that come out of training.
Put it to work
AI agents that look after your customers
From €49/mo + VAT
Voice, WhatsApp and email with the phone line included: agent, numbers and CRM in one fee.
See plans and pricingMost models described as open release the weights and the code to run them, but not the training data. So a distinction is increasingly drawn between open weights and open source in the strict sense. In 2024 the Open Source Initiative published a definition of open-source AI that requires, among other things, enough information about the data to be able to reproduce the model.
Families such as Llama, Mistral, Qwen, Gemma and DeepSeek have released models with downloadable weights, each under its own licence and conditions.
Related to this term
UTIO agents
- Available 24/7 in 26 languages
- Set it up yourself for €0: built from your website
- 14-day free trial, no card needed
From €49/mo + VAT
See plans and pricingHow it works
Using an open model usually goes like this:
- Choose a model based on task, language, size and licence.
- Download the weights, often in a slimmed-down version produced by quantization so it fits in less memory.
- Run it on your own server, in the cloud of your choice or on a powerful workstation.
- Adapt it if needed, with fine-tuning or with RAG over your documents.
The main advantage is control: you decide where it runs and what data it sees. The downside is that maintenance, security and the hardware bill become yours.
Example
A small health insurer in Bath wants to sort the queries it receives from members. Because the data is sensitive, its data protection officer would rather the text never left its own infrastructure.
The technical team tries an open-weight small language model under a licence that allows commercial use, quantized to fit its server, and tunes it on old, anonymised queries. It works, but they discover the hidden cost: someone has to update it, monitor it and respond if it falls over on a Saturday.
How it affects your business
An open model makes sense if you need full control over where data is processed, if your volume justifies the infrastructure, or if you want to customise it deeply. For most small businesses, the practical option is a service that already runs the model and gives you contractual safeguards.
At UTIO, the agents’ language model runs on our own servers, with fallback to European and US platforms under contractual safeguards, and recordings, transcripts and the CRM are stored in Spain. That way an AI agent, such as the voice agent, works without you having to maintain any model.
Frequently asked questions
What's the difference between open source and open weights?
Open weights means you can download the trained model and run it. Open source in the strict sense also requires being able to study and reproduce it: enough information about the training data, the code, and a licence that lets you use and modify it for any purpose.
Is an open model free?
The licence may be, but running it isn't: you need servers with graphics cards, maintenance and someone to look after it. At low volumes, paying a provider per use usually works out cheaper than building your own infrastructure.
Can I use an open model commercially?
It depends on the licence. Some, such as Apache 2.0 or MIT, allow it with few conditions. Others impose restrictions on use or on company size. Read each model's licence before building it into a product.
Related terms
- LLM
An LLM (large language model) is an AI model trained on vast amounts of text to predict the next word. That single skill lets it draft, summarise, translate, classify and hold a conversation.
- Small language model (SLM)
A small language model (SLM) is a language model with far fewer parameters than an LLM. It knows less about everything, but it replies faster, costs less and can run on a modest server, a laptop or even a phone.
- Quantization
Quantization is a technique that stores an AI model's parameters as lower-precision numbers, for example 8 or 4 bits instead of 16. The model takes up less memory and often runs faster, in exchange for a loss of quality that can be small.
- Fine-tuning
Fine-tuning means further training an AI model that's already been pre-trained, using a small set of your own examples, so it specialises in a particular task, format or style without starting from scratch.
- EU AI Act
The EU AI Act is Regulation (EU) 2024/1689, the European Union's artificial intelligence law. It sorts AI systems by level of risk and sets obligations for those who build and use them, from banning some uses outright to transparency duties.
Updated on