Models & Architectures

Open-Weight Model in plain English.

Also known as: open weights,open-weights AI

The one-sentence version

A model whose trained parameters are published for anyone to download and run, even if the training data and code are not.

An open-weight model is one whose trained parameters — the billions of numbers that make it work — are released publicly, so anyone can download, run, fine-tune, and deploy it. This is narrower than "open source": the weights are open but the training data, the full training code, and sometimes commercial use may be restricted by licence. Llama, Qwen, DeepSeek, Gemma, Mistral, and GPT-OSS are open-weight; ChatGPT, Claude, and Gemini are not. Open weights matter because they let companies run models on their own infrastructure for privacy or cost reasons, let researchers inspect and modify them, and put a price floor under the whole market — a closed model has to be meaningfully better than the best free one to justify its fee. Their risk is that safeguards can be removed by whoever runs them.

Read the full guide