loading
loading
A model whose trained weights you can download and run yourself, versus a closed model you only reach through an API.
Open-weights (Llama, Mistral, Qwen, DeepSeek) means the parameters are public, so you can self-host, fine-tune with LoRA, quantize, and run offline with no per-token bill or vendor lock-in. It's not the same as 'open source' — you usually get weights but not the training data or full recipe, and licenses can restrict commercial use. The tradeoff versus a closed frontier API like Claude: you own the stack and the privacy, but you also own the GPUs, the ops, and the gap in raw capability.
Plainly
Think of Open-weights as the brain part that guesses or decides. A model whose trained weights you can download and run yourself, versus a closed model you only reach through an API.
In practice
Use it when model choice, prompts, latency, cost, or quality affect the product result. In practice, define the owner, input, output, and failure mode before you rely on it.