What Is an Open-Weight Model?
Also called open-source model, open weights, open model
Short Answer
An open-weight model is a language model whose trained weights are published so anyone can download it and run it on their own computers or cloud. Llama, Mistral, Qwen, DeepSeek and gpt-oss are examples. It differs from a closed model, which you can only use through the provider's API.
- Taught in lesson 8
Why It Matters
Open-weight models give organisations more control: they can run a model on their own servers, keep sensitive data inside, adapt it with fine-tuning and avoid depending on one provider. Many companies, especially in regulated fields, weigh open-weight against closed models for each use case.
How It Works
People often call these models open source, but most publish only the weights, not the training data or full training code, and some licenses restrict commercial use. Open-weight is the more precise term.
Companies choose open-weight models to keep sensitive data on their own infrastructure, to control cost at large volumes, or to fine-tune the model for a narrow task. The trade-off is that they must host, secure and update it themselves.
Running an open-weight model means taking on work a provider would otherwise do: choosing hardware, serving the model efficiently, updating it and securing it. The trade-off is control over data, cost at high volume and the freedom to fine-tune. Licences differ: some allow almost any use, others limit commercial use or very large companies, so the licence must be checked before building a product.
Open-Weight Model in Practice
Everyday Example
A developer runs a small open-weight model on a laptop with a tool like Ollama to draft text without sending anything to the internet.
Industry Example: Pharma
A hospital research team runs an open-weight model inside its own secure environment so patient records never leave its servers, which helps with data protection rules.
Where You Meet It
- Model hubs such as Hugging Face.
- Companies running AI on their own servers for data protection.
- Fine-tuning projects.
- Discussions of AI regulation and openness.
Key Ideas Around It
- Weights
- The learned numbers that make up a trained model.
- Closed model
- A model used only through a provider's service.
- Self-hosting
- Running a model on your own infrastructure.
- Model licence
- The terms that say how a model may be used.
A Common Mix-Up
Open-weight is not the same as open source. Many models publish weights but not their training data or code, and their licences may restrict use.
Check YourselfWhy might a hospital choose an open-weight model?
To run it on its own servers so patient data never leaves its control, and to adapt it to its needs.
Learn It on ai2hired
S0 · #8Tokens, Context Windows and Model Families
Coming in Season 0, Start Here. Every ai2hired lesson is free on YouTube.
Questions People Ask
Are open-weight models free?
The weights are often free to download, but running them needs hardware or cloud capacity, and the license may set conditions on use.
Are open-weight models as good as closed models?
The best closed models usually lead, but open-weight models have narrowed the gap and are often good enough for focused tasks.
Related Terms
Learn AI From Zero to Hired
Free lessons on YouTube, each one built from what real AI job ads ask for.