Open weights
You can download the model itself and run it yourself (Llama, Mistral, Qwen).
Last updated
Analogy
Open source for models: you get the build output, not just an API key.
In an agent
Open weights are what make self-hosted, private agents possible.
Practice
Flashcards
- Interview: when do you pick an open-source model?High-volume narrow tasks (embeddings, classification), strict privacy, or when cost per call matters.
Interview questions
- When would you use an open-source model?
- High-volume narrow tasks like embeddings and classification.
- Strict privacy requirements, or when cost per call matters.
- In DevPilot, Hugging Face models do embeddings and PR classification locally in Node.
Questions people ask
What are open-weight models?
Open-weight models, such as Llama, Mistral and Qwen, let you download the trained model files and run them yourself. You get the finished model, not just an API key, though not always the training data or code. Licenses vary, so check each one before shipping.
Why do open weights matter for agents?
They make private, self-hosted agents possible. With open weights, a company can run the model inside its own network, so sensitive data never reaches an outside API. They also allow fine-tuning and quantisation for narrow tasks, which can make small agents fast and cheap.