OLMo 3: Defining the Future of Open-Science AI with Model Flows

Join us for a talk by Hannaneh Hajishirzi, Professor at UW & Senior Director of AI at Ai2. This talk is part of the Kempner Seminar Series, a research-level seminar series that covers topics related to the basis of intelligence in natural and artificial systems.

In this talk, I introduce Olmo 3, a family of state-of-the-art, fully-open language and thinking models at the 7B and 32B parameter scales with a diverse set of capabilities, including long-context reasoning, function calling, coding, instruction following, general chat, and knowledge recall. With the Olmo 3 release, we go further by providing complete access to its entire model flow—the full lifecycle of a language model, including every stage, checkpoint, datapoint, and dependency required to create it. This enables infinite customization through intervention at any stage of the model development process—not just the final weights. 

In particular, I describe how we train Olmo 3 Think to perform step-by-step reasoning by generating intermediate thinking traces before producing a final answer.  Olmo 3 Think is the strongest fully-open thinking model, narrowing the gap to the best open-weight models of similar scale, such as the Qwen 3 32B thinking on our suite of reasoning benchmarks, while being trained on six times fewer tokens.

Hanna Hajishirzi is a Professor of Computer Science at the University of Washington and a Senior Director of AI at AI2. Her research spans generative AI and natural language processing, with a focus on building pioneering, open-science AI solutions. She co-leads the OLMo and Tulu projects, advancing fully open language and reasoning models to accelerate the science of AI, empower the research community, and champion openness as a driver of innovation. These models have been downloaded more than 10 million times as of 2025 and were recognized with GeekWire’s Innovation of the Year award. She is a co-PI of a $152M NSF- and NVIDIA-supported grant to develop the next generation of open multimodal models.

She is a recipient of the ACL Fellowship (2025), the Sloan Fellowship (2021), the Uncommon Thinker Award (2025), the NSF CAREER Award (2021), Torode family Career development professorship (2022), the Allen Distinguished Investigator Award (2014), the UIUC Alumni Award (2024), and was a finalist for the VentureBeat Women in AI Award (2024, 2025). Her research has earned recognition at leading venues, with papers receiving or being finalists for awards at EMNLP 2025, ACL 2025, CVPR 2025, ACL 2024 (Best Paper and Best Resource Paper), CVPR 2022, AKBC 2020, and SIGDIAL 2012.

Her work has been widely featured in leading magazines and newspapers. She has delivered keynote talks at premier venues including the White House, EMNLP, COLM, PyTorch, and Linux.