In plain English
Open-source AI refers to models or tools whose code, weights, or architecture are publicly available for anyone to use, modify, and distribute, subject to their licence terms. This contrasts with closed models that are only accessible through a provider API. Open models give organisations more control, including the option to run them on their own infrastructure.
Why it matters
Open models can be run privately, which matters for data sensitivity, cost control, and avoiding lock-in to a single provider, though they shift more responsibility onto you.
A worked example
Meta's Llama models can be downloaded and run on your own servers, unlike a closed model you can only reach through a paid API.
Common confusion
Open source does not always mean free for any use. Licences vary, and some restrict commercial use or redistribution, so the terms need checking.