Open-weight model
In shortAn open-weight model is an AI model whose trained weights are published for download, so you can run and often adapt it yourself.
In plain words
A trained model is a very large file of numbers, the weights or parameters. With a closed model, you reach it only through the provider's service. With an open-weight model, the provider releases the weights, so you can download them, run the model on your own hardware and, in many cases, adapt it with fine-tuning.
Open weights are not the same as open source. The Open Source Initiative has published an Open Source AI Definition. It requires freedoms to use, study, modify and share the system, and it asks for three parts in the form needed to modify it: information about the training data, the code used to train and run the system, and the parameters. Many open-weight releases give you the weights but not the full training data or code, and they come with their own licence. So an open-weight model can be free to download and still limit how you may use it.
Why it matters
- Control and privacy: Run the model on your own machine or server, so prompts need not leave your network. See local models.
- Cost: No per-token fee from a provider, but you pay for hardware, power and your own time.
- Customisation: You can tune the model to your data and tasks.
- Licence: Read it. Some licences limit commercial use, set size thresholds for large companies, or forbid certain uses. Model cards on the Hugging Face Hub describe a model's licence, intended use and limits.
- Responsibility: You handle updates, safety filters, hosting and fixes. A provider's service does this for you.
- Trust: Download weights only from sources you trust, because files from unknown places can be tampered with.
Example
A small firm wants to summarise internal documents without sending them to a cloud service. It reads the model card of an open-weight model, checks that the licence allows business use, downloads the weights from the Hugging Face Hub and runs the model on a server in its own office. It pays for the server, not for each token.
Common misconceptions
- "Open weights means open source." By the OSI definition, it often does not.
- "Free to download means free to use any way." The licence sets the rules.
- "I can read the weights, so I can see how it thinks." The weights are billions of numbers, not readable rules.
Questions and answers
What is the difference between open-weight and open-source AI?
Open-weight means the trained weights are published. Open-source AI, by the OSI definition, also needs information about the training data and the code, plus freedoms to use, study, modify and share.
Can I use an open-weight model commercially?
It depends on the licence. Read it before you build on the model. Some allow business use freely, others set limits.
Do I need powerful hardware to run one?
It depends on the model size. Small models run on a laptop. Large ones need servers with much memory.
Read next
- Local model
A local model is an AI model that runs on your own computer or server instead of a provider's cloud, so prompts need not leave your device.
- Fine-tuning
Fine-tuning trains an existing model further on your own examples, so it follows a style, format or task more reliably.
- Parameters
Parameters are the numbers, also called weights, that a model adjusts during training and then uses to answer; their count is a rough measure of size.
- Large language model (LLM)
A large language model (LLM) is an AI system trained on huge amounts of text to predict the next token, which lets it write, summarise and answer.
- Inference
Inference is the act of running a trained AI model to produce an output from your input, such as an answer to a prompt.