According to the Open Source Initiative, open source AI refers to “an AI system that is made available under terms that allow users to freely use the system for any purpose, study how it works, inspect its components, modify it and share it—whether or not the system is changed.” This approach emphasizes accessibility, flexibility and transparency, allowing anyone to collaborate on improving the technology.
Open source AI contributes to creating a more democratic and innovative ecosystem. Making AI systems available to the public fosters greater collaboration and accelerates progress in the field. Major organizations, such as IBM (Granite™), Meta (Llama) and Mistral AI, are leading efforts to develop open source AI, providing sophisticated tools to developers and researchers worldwide. For instance, IBM’s AI platform, watsonx.ai™, uses several key AI open source tools and technologies to support innovation and performance.
Large language models (LLMs), advanced AI systems that use deep learning and vast datasets to generate text, convert languages and create various forms of content, can be divided into 2 main categories: proprietary LLMs and open source LLMs. Both types are key to generative AI, a technology that produces new content based on learned patterns and data—text, images or music.
Open source LLMs, in particular, are vital to the generative AI ecosystem because they promote a more transparent, accessible and community-driven approach compared to proprietary models.
Compared to proprietary LLM models, such as Google’s LaMDA and OpenAI’s ChatGPT-3 and GPT-4 , open source LLMs offer distinct benefits. For instance, they allow developers to inspect, modify and improve the models, enabling rapid innovation and customization.
As another example, IBM® Granite™ AI models, available under Apache 2.0 licenses on Hugging Face and GitHub, deliver comparable performance to larger systems but require far fewer computing resources, showcasing how open source models can provide more efficient alternatives in the generative AI space. Moreover, open source AI provides a cost-effective solution for organizations seeking to fine-tune their generative AI models with proprietary data.
In 2025, the Chinese company DeepSeek disrupted the AI markets with R1, its large language model that costs just USD 5.6 million to train, a fraction compared to commercial leaders such as ChatGPT. This development demonstrated how open source LLMs can compete with commercial offerings, in this case, using more affordable hardware and fewer advanced microchips.5 This shift underscores how open source models contribute to generative AI technology’s democratization.
link