What's the difference between closed, open‑source and open-weight AI? A researcher explains

Closed models are proprietary and accessible only via API, while open-source models allow full access to code, data, and weights, enabling modification and redistribution. Open-weight models fall in between, providing the trained parameters for download but not necessarily the training data or code to modify the architecture.

The article notes that Meta released LLaMa as an open-source large language model on Feb. 24, 2023, which was later criticized by some for not being fully open-source due to restrictions on training data access. The researcher quoted in the piece emphasizes that the terms are often misused, with 'open-weight' becoming a marketing term rather than a precise technical classification.

The piece concludes that understanding these distinctions is important for evaluating AI transparency, security, and ethical implications. It also notes that the lack of standardized definitions can lead to confusion among policymakers and the public, potentially hindering informed decision-making around AI governance.