What is an open-weight model?
Updated · By Robert Breen
An open-weight model is an AI model whose trained parameters, called weights, are released publicly so anyone can download it and run it on their own computer or server. Closed models, like the ones behind ChatGPT, are only available through the provider's app or API.
Why it matters for a small business
Running a model yourself changes who sees your data. With an open-weight model on your own machine or private server, prompts never leave your control, which matters for firms with strict confidentiality rules. You also avoid per-request API charges, though you pay instead in hardware, hosting and someone's time to keep it running.
The tradeoffs are real for a small business. The most capable closed models are often ahead on quality, setup takes technical skill, and you become responsible for updates and security. For most first automations, an API model is simpler. Open weights become worth a look when privacy rules, volume or a need to avoid a single vendor push you there.
In a real lesson: Build an AI Customer Follow-Up Agent for Plumbers
Stepthrough's lessons use OpenAI's hosted models, so there is no lesson where you run an open-weight model. The closest thing you can practice is the slot where one would plug in. In the plumbing follow-up agent lesson, you build an agent for Blue Maple Plumbing, a made-up residential plumber, that writes short follow-up texts after service calls.
The model is a separate piece hanging off the AI Agent. You click the + under Chat Model, choose OpenAI Chat Model and pick gpt-5-mini. The system prompt (thank the customer by first name, one care tip per job, under 60 words, never invent prices or warranty terms), the Simple Memory and the Google Sheets Tool do not depend on which model sits in that slot.
n8n also offers chat model nodes for other providers, including one for Ollama, a popular way to run open-weight models on your own hardware. Swapping the model node is the easy part; the harder part is hosting the model and confirming its follow-ups for jobs like water heater replacements are as good as what you had.
Try this lesson free or read the step-by-step guide.
Common confusions
Open-weight vs open source
Open weights means the trained model is downloadable. Fully open source would also mean the training code and data are available, under an open license. Many models called "open" share weights but not training data, and their licenses can carry conditions.
Open-weight vs free
The download may cost nothing, but running a capable model needs powerful hardware or paid cloud servers. Compare that with API costs for your actual volume before switching.
Tips
- Read the model's license before using it commercially.
- Test an open-weight model on your real tasks side by side with your current model before relying on it.
Related terms
Where to learn more
Frequently asked questions
- What are examples of open-weight models?
- Well-known families include Meta's Llama models, several models from Mistral, and OpenAI's gpt-oss models. Each has its own license terms.
- Can I use an open-weight model in n8n?
- Yes. n8n has chat model nodes for several providers, including Ollama for models you run yourself, and you attach one to the AI Agent the same way as the OpenAI Chat Model.