Open-weight vs closed models
Models you can download and run yourself vs models you use only through a company's service.
What is it?
A closed (proprietary) model is only available through its developer's apps or API. You send your request to the company's servers and get the answer back, but you cannot download the model itself. An open-weight model has its trained parameters published, so anyone allowed by its license can download and run it on their own computer or server.
"Open-weight" is not always the same as fully open source: the training data and code may not be published, and licenses can include restrictions. Read the license before using a model commercially.
Which models use it?
- OpenAI's GPT models, Anthropic's Claude models and Google's Gemini models are mainly offered as closed, hosted services.
- Meta's Llama models are published with downloadable weights under Meta's license.
- Mistral AI publishes some open-weight models and also offers hosted models.
- Google's Gemma models are open-weight models related to Gemini research.
- Hugging Face hosts many open-weight models that people can download.
How to use it
- Decide what matters most for your task: ease of use, quality, privacy, cost or control.
- For quick everyday use, start with a hosted assistant or API.
- If data must stay on your own systems, look at open-weight models you can run locally or on your own servers.
- Read the model's license to confirm what uses are allowed.
- Test a few candidate models on your real tasks before committing.
- Plan for maintenance: self-hosting means you handle updates, security and hardware.
Benefits
- Open-weight models can run offline and keep data on your own hardware.
- Open-weight models can be customized and fine-tuned more freely.
- Closed models are usually easier to start with and need no hardware of your own.
- Having both options reduces dependence on a single company.
Limits and risks
- Running large open-weight models needs capable hardware and technical skill.
- Closed models can change or be retired by the provider without your control.
- Licenses differ, and some restrict commercial or large-scale use.
- Open-weight models can also be misused, and safety controls are up to whoever runs them.
Prompt template
Copy it, replace the parts in [brackets] and paste it into any AI chat.
Help me choose between an open-weight model and a hosted closed model for [project]. My priorities are [privacy / cost / quality / offline use]. My technical skills are [describe]. Compare the two options in a short table covering setup effort, data privacy, cost structure, customization and maintenance. Then recommend one option and explain what I should test first. Tell me what I should verify in the official documentation or license.
More in Foundations
- Large language models (LLMs)The text-predicting systems behind chat assistants: what they are and what they can do.
- Transformers and attentionThe neural network design that most modern language, image and speech models are built on.
- Tokens and context windowsHow models measure text, and why there is a limit to how much they can read at once.
- Training vs inferenceThe difference between teaching a model and using it, and why it matters to you.
- Choosing the right model (cost, speed, quality)How to match a model to your task instead of always using the biggest one.