Open weight
Open weight is an AI model that makes the model's weight file public so that anyone can download and use it. Representative examples include Llama and DeepSeek.
Open weight is a method that discloses the weights, which are the core of the AI model, that is, the numerical value file obtained through learning, so that anyone can download it and run the model directly on their computer or server. It can be compared to receiving the whole dough instead of buying the finished bread, baking it at home, and modifying it to your liking.
Unlike closed models that can only be used through APIs, open-weight models can be operated without data being leaked or freely fine-tuned, making them popular with companies and researchers. Llama, DeepSeek, and Mistral models have led this trend and narrowed the performance gap with closed models.
Open weight and open source are often used interchangeably, but are strictly different concepts. In most cases, only the weights are public and the training data and training code are private, so it is often impossible to reproduce the model from scratch.
✅ Why it matters
- You can protect data sovereignty and security by running it on yits own infrastructure
- Free modifications such as fine tuning and lightweighting are possible
- Can be operated without API usage fees, reducing costs when used on a large scale
⚠️ Limits and debates
- It is difficult to reproduce or fully validate the learning process using weights alone
- There is a risk of variant models being distributed with safeguards removed
- Requires significant hardware and expertise to run and operate