On Tuesday, AI2, the nonprofit AI research organization founded by Microsoft co-founder Paul Allen, introduced OLMo 2, the latest addition to its Open Language Model series. Unlike many so-called “open” language models, such as Meta’s Llama, OLMo 2 fully adheres to the Open Source Initiative’s definition of open-source AI. This means all tools and data used in its development are freely accessible to the public.
The Open Source Initiative, which finalized its open AI definition in October, aims to ensure transparency and accessibility in AI development. Ai2’s OLMo series has consistently met these standards since its first release in February.
According to AI2, OLMo 2 was built using open training data, transparent evaluations, reproducible training recipes, and intermediate checkpoints. “By openly sharing our data, recipes, and findings, we aim to empower the open-source community to explore new and innovative approaches,” the organization stated in a blog post.
The OLMo 2 family includes two models: OLMo 7B with 7 billion parameters and OLMo 13B with 13 billion parameters. Parameters measure a model’s ability to handle complex tasks, and larger models tend to deliver higher performance. Like other language models, OLMo 2 can handle tasks such as answering questions, summarizing documents, and generating code.
Ai2 trained these models on a data set comprising 5 trillion tokens, equivalent to 750 billion words. This extensive dataset included high-quality websites, academic papers, Q&A forums, and math workbooks, both human-generated and synthetic.
Performance-wise, OLMo 2 models demonstrate significant improvements over earlier iterations, with the OLMo 2 7B even surpassing Meta’s Llama 3.1 8B. Ai2 describes OLMo 2 as “the best fully open language models to date.” These models and their components are available under the Apache 2.0 license, making them suitable for commercial use.
While the rise of open models has sparked concerns about potential misuse—such as reports of Llama models being used for defense tools in China—Ai2 remains committed to open innovation. Ai2 engineer Dirk Groeneveld emphasized that the benefits of open models, including promoting ethical advancements and equitable access, outweigh the risks.
