As the open source ecosystem continues to evolve, artificial intelligence (AI) has become a core area of innovation and collaboration. Within this landscape, it is critical to establish clear definitions for “open source AI models” and “open science AI models.” While the Open Source Initiative (OSI) is diligently working on defining the term “Open Source AI,” our work focuses on a narrower scope, extending from the Model Openness Framework we’ve developed in LF AI & Data. These definitions represent a natural evolution of our ongoing efforts and are aligned with the broader goals of openness, transparency, and collaboration that underpin the open source community.
In this post, I’ll outline two key definitions—“open source AI models” and “open science AI models”—and present the case for their adoption within the community. By establishing clarity around these terms, we can foster greater collaboration, innovation, and trust across the AI landscape. In particular, community model projects within the LF AI & Data community will be expected to meet the requirements of the Open Source AI Model requirements below.
The Need for Standardized Definitions in AI Models
Clear and standardized definitions are helpful for ensuring consistent understanding and collaboration across our community and the open source ecosystem. Without such standards, terms like “open source AI model” and “open science AI model” can become ambiguous, leading to confusion and fragmented approaches. By adopting standardized definitions, we create a common language that everyone—whether they are researchers, developers, or executives—can align around and set appropriate expectations. This is particularly important in open source AI model communities, where collaboration is a key driver of innovation. When stakeholders have different interpretations of what it means for a model to be “open source” or “open science,” it leads to misaligned expectations and inconsistencies in how these models are built, shared, and used. Ultimately, standardized definitions create a more cohesive and productive ecosystem. They enable us to move forward with confidence, knowing that we’re all working from the same playbook, which is vital for building sustainable, transparent, and widely adopted AI technologies.
Open Source AI Models
“Open source AI models” are built on the foundational principles of open source software. By making the key components available and reusable, these models empower developers, researchers, and organizations to innovate and contribute to the AI field at a faster pace, reducing redundancies and promoting collaboration across industries.
Open source artificial intelligence (AI) models enable anyone to reuse and improve an AI model. “Open source AI models” include the model architecture (in source code format), model weights and parameters, and information about the data used to train the model that are collectively published under licenses, allowing any recipient, without restriction, to use, study, distribute, sell, copy, create derivative works of, and make modifications to, the licensed artifacts or modified versions thereof.
Open source AI models provide an open playing field where developers, researchers, and enterprises can innovate without being locked into proprietary systems. This accelerates the rate of technological advancements, as seen in many other domains of open source software.
Open Science AI Models
Open science AI models take the concept of open source AI a step further by adding an additional layer of transparency and reproducibility.
Open science AI models enable anyone to reuse and improve an AI model, while providing additional transparency and reproducibility. “Open science AI models” are open source AI models that additionally include licenses to all source code, training data, configuration files, research, and documentation required to reproduce a similar AI model without restriction.
Open science AI models offer an added layer of transparency. By making the entire process — from data collection to model creation — open and accessible, these models provide a way for the community to build AI systems that are more understandable and accountable. They enable the reproducibility of AI research, a critical aspect that is often overlooked. By providing all the necessary components to replicate AI models, researchers and developers can validate findings and develop novel methods that advance AI systems research, leading to more robust and trustworthy AI systems.
This broader scope ensures that others can not only use and modify the model but also fully understand and replicate its development process. Open science AI models are aligned with the core values of the open science movement, promoting transparency, reproducibility, and scientific rigor in AI research. By embracing these models, the community gains the ability to scrutinize AI developments more closely, thereby fostering trust and enabling the creation of more reliable, unbiased, and trustworthy AI systems.
A Natural Evolution of the Model Openness Framework
The definitions are an extension of the Model Openness Framework, which has served as a guiding principle for how we think about transparency and openness in AI. While OSI’s work on defining “Open Source AI” is invaluable, our focus on these specific definitions is much narrower in scope, addressing the unique challenges and opportunities that arise from making AI models open and innovating with them in a community context.
Moving Forward Together
We recognize that defining what constitutes an open source AI model and an open science AI model is a complex task with many facets. However, by adopting these definitions, we can create a shared understanding within the community and move toward a more open, transparent, and collaborative AI future.
A shared understanding ensures that when we refer to an “open source AI model,” we are all speaking about the same core principles: the availability of model architecture, weights, parameters, and information about the data used to train the model under licenses that allow for unrestricted reuse and modification. Similarly, when we refer to an “open science AI model,” we are collectively referring to models that not only adhere to open source principles but also provide the transparency and documentation needed for full reproducibility.
I invite you to reflect on these definitions and consider how they might align with your work in the AI and open source ecosystems. Your feedback and participation are crucial as we continue to refine and evolve these concepts to better serve the needs of the community.
Let’s work together to build a future where open source and open science AI models can thrive, and where transparency and collaboration are at the heart of AI innovation.
To participate in this effort, please join the Generative AI Commons and the LF AI & Data Technical Advisory Council (TAC).
