# What is an open-weight model, and is it the same as open source AI?

A model whose trained weights anyone can download and run on their own hardware, under a licence that sets what you may do with them. Open weights alone do not make a model open source.

- Page: https://tecyfy.com/concepts/open-weight-model
- Subject: How AI works (https://tecyfy.com/concepts)
- Level: beginner
- Also called: Open weights, Open-weights model, Open model

A trained model is an architecture plus its [weights](https://tecyfy.com/concepts#weight), the billions of numbers learned in training that turn an input into an output. When a maker publishes those weights for download, anyone can run the model on their own machines, [fine-tune](https://tecyfy.com/concepts/fine-tuning) it on their own data, and keep every prompt inside their own network. That is an open-weight model. A closed model, by contrast, is reached only through its maker's API or apps.

## Open weights are not open source

The Open Source Initiative's definition of open source AI asks for 3 things, each under terms that let anyone use, study, change and share them: the parameters (the weights), the complete code used to train and run the model, and enough information about the training data that a skilled person could build a substantially equivalent system. A model that publishes only its weights meets the first and not the other 2. You can run it and change it, but you cannot see how it was made or rebuild it.

So "open" in a model's name tells you little. Read 2 things before you build on one.

## Read the licence

Each open-weight model comes with its own licence, and the conditions differ from model to model. Mistral Large 4's licence, for example, is a modified MIT licence that lets anyone use, change, fine-tune and sell with the model for free, except that a business running a Model-as-a-Service offering, an API that lets customers send their own prompts and settings to the model, needs a separate commercial licence from Mistral. Internal use is exempt. The same model can be free for your product and not free for a hosting company.

## Count the hardware

Open weights move the cost from the maker's bill to yours. A large model needs many GPUs: Mistral Large 4's weights alone are about 1,061 GB in [FP8](https://tecyfy.com/concepts/fp8), and Mistral's own serving examples use 8 GPUs. [Quantized](https://tecyfy.com/concepts/quantization) versions shrink that, and serving engines such as [vLLM](https://tecyfy.com/concepts/vllm) get more answers out of each GPU, so whether running it yourself beats paying per token depends on how steady your traffic is and whether your data may leave your network at all. Mistral's model card itself recommends its API when local serving falls short.

Our [GPU memory calculator](/ai-agents/gpu-memory-calculator) works out the memory an open model needs on a given GPU, and [vLLM vs Ollama](/blog/vllm-vs-ollama) compares 2 ways to run one.

## Related

- [Weight](https://tecyfy.com/concepts#weight): A trainable parameter in a neural network that scales input values and is updated during training.
- [Parameter](https://tecyfy.com/concepts/parameter): A number inside a model that training sets, a weight or a bias. "70B" in a model's name means about 70 billion of them.
- [Large Language Model](https://tecyfy.com/concepts/large-language-model): A model trained on a huge amount of text to predict the next token. Further training turns it into an assistant that follows instructions.
- [Fine-Tuning](https://tecyfy.com/concepts/fine-tuning): Training an existing model a little further on your own examples so it picks up a task, a format or a style. Often a good prompt does the job first.
- [vLLM](https://tecyfy.com/concepts/vllm): An open-source engine for serving language models fast, built on PagedAttention, that answers requests through the same kinds of API the big model makers offer.
- [Model Card](https://tecyfy.com/concepts/model-card): The documentation that comes with a trained model, saying what it is for, how it was trained and evaluated, and where it should not be used.
- [FP8](https://tecyfy.com/concepts/fp8): An 8-bit floating-point number format, in 2 kinds, E4M3 and E5M2. It halves a model's memory against BF16, and in tests it matched 16-bit results.

## Sources

- [The Open Source AI Definition 1.0](https://opensource.org/ai/open-source-ai-definition)
- [Mistral Ecosystem Open-Weight License](https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B/blob/main/LICENSE)
- [Mistral Large 4 - Le Chonk (model card)](https://huggingface.co/mistralai/Mistral-Large-4-1T-A52B)