Open-weight model

An open-weight model is an AI ✕ Artificial intelligence is a computer system’s ability to perform tasks we associate with human intelligence, such as understanding language, recognizing images or making decisions. More in the glossary → model — usually a large language model ✕ An LLM is an AI model trained on huge amounts of text to understand and generate language. It is the engine behind assistants like ChatGPT. More in the glossary → — whose "weights", that is, the numeric parameters that result from training, are released openly. This lets anyone download it, run it on their own servers, fine-tune it to their needs and build it into their products, without depending on a single company’s cloud ✕ "The cloud" refers to services and storage that run not on your device but on external servers on the internet. You can access them from anywhere. More in the glossary → . It contrasts with closed or proprietary models, such as those from OpenAI, which can only be reached remotely through an API ✕ An API is a set of rules that lets two programs talk to each other. It works like a go-between that carries requests from one application to another and returns the response. More in the glossary → . It should not be confused with " open source ✕ Open source is software whose source code is public: anyone can view, use, modify and share it. It favours transparency and avoids depending on a single vendor. More in the glossary → " in the strict sense: the weights being public does not mean the training data ✕ Training data is the set of examples an AI model learns from. Its amount and quality determine how good —and how biased— the model will be. More in the glossary → and code are too, which is why the term "open-weight" is preferred. Well-known examples are Meta’s Llama, Mistral, DeepSeek or Google’s Gemma. Its advantages are privacy, control, lower cost and the possibility of customisation, and it has fuelled a very active community around AI you can self-host.