Olmo 3: Charting a path through the model flow to lead open-source AI

Allen Institute for AI (AI2) has released Olmo 3, a new family of open-source AI models that includes the entire model flow, marking a significant step toward transparent and collaborative AI development. Unlike traditional models treated as isolated snapshots, Olmo 3 offers full access to every stage, checkpoint, dataset, and dependency involved in its creation. This comprehensive approach allows researchers and developers to modify, adapt, and extend the model's capabilities effectively by integrating domain-specific knowledge deep within the development pipeline.

The Olmo 3 family features models like Olmo 3-Think (32B), which enables inspection of intermediate reasoning traces, and Olmo 3-Base (7B, 32B), which delivers top-tier performance among fully open base models. Olmo 3-Instruct (7B) excels in chat and tool use, while Olmo 3-RL Zero (7B) provides a reinforcement learning pathway for complex reasoning behaviors. By releasing all components under permissive open-source licenses, AI2 aims to foster innovation and customization within the AI community, empowering users to build their own systems on a robust and transparent foundation.

New Olmo 3 models provide the open-source community with state-of-the-art AI and complete lifecycle traceability, fostering trust and collaboration.

RecursivAI

AI2 Unveils Olmo 3: Full Access to Model Flow