What is Open weights?

Models whose trained parameters are publicly released, letting you run or self-host them. Examples include Llama and DeepSeek.

Open-weights models are AI models whose trained parameters (the weights) are published for anyone to download. That means you can run the model on your own hardware or cloud servers, fine-tune it on your data, quantize it to fit smaller machines, and keep all inputs and outputs inside your own infrastructure. Examples include Meta's Llama family, DeepSeek, Mistral, and Qwen. Open weights is not the same as open source: the training data and code are often not released, and licenses vary, with some restricting commercial use above certain scales. The trade-off versus closed API models is control and privacy versus convenience: you take on hosting, GPUs, and updates yourself, or pay a hosting provider to serve the open model for you, often at lower per-token prices than frontier closed models.

Example

A healthcare company self-hosts Llama on its own servers so patient data never leaves its network, and fine-tunes the model on internal clinical documentation.

Why it matters

Open weights give you data privacy, customization, and freedom from vendor lock-in, usually at some cost in raw capability and convenience. If compliance or cost at scale drives your decision, they belong on your shortlist. Browse the AI tools directory or the model leaderboard to put it into practice.