What is an open-weight model?
An open-weight model is an AI model whose trained neural network "weights" are published, so anyone can download it, run it in their own environment, or fine-tune it. Weights are the bundle of core values the model acquired through training.
Because the weights are open, you can run the model locally without an internet connection or train it further for a specific purpose.
What is the difference between open-weight models and open-source AI?
Fully open-source AI releases not only the weights but also the training data and training code, so it can be reproduced from scratch. Open-weight, by contrast, often releases only the weights, the end product, and keeps the data and code private.
So open-weight is closer to a subset of open source, and it is important to recognize that the two concepts are not always the same.
How do you run an open-weight model?
You can download the published weights and run them directly on a personal PC or server, or put them on a cloud model-hosting service. The memory you need (VRAM/RAM) depends on the model size.
Before downloading, always check the license terms for whether redistribution and commercial use are allowed. Published weights do not mean commercial use is free.
How do you compare open-weight model rankings and performance?
Which model is "number one" keeps changing with timing and evaluation criteria, so instead of memorizing a fixed ranking, it is more accurate to check public leaderboards and benchmarks directly.
When comparing, weigh not only performance scores but also model size, the hardware needed to run it, and license restrictions to make a useful adoption decision.