Research institute · Research orgs · Seattle, WA · Est. 2014
The Allen Institute for Artificial Intelligence (Ai2) is a Seattle-based, nonprofit AI research institute founded in 2014 by Paul Allen, focused on building “breakthrough AI” for major societal challenges through large-scale open models, open data/evaluations, robotics, and AI “for science” and “for the planet.” Ai2’s technical strategy emphasizes open, end-to-end releases (model weights, training data/artifacts, evaluations, and code) so the research community can inspect, reproduce, and build on its work.
This openness is reflected across its product lines: open language/multimodal model families (for example, Olmo and Molmo), Earth observation and conservation platforms (for example, OlmoEarth, EarthRanger, and Skylight), and embodied/robotics systems trained via simulation-first approaches. Institutionally, Ai2 positions itself as an “infrastructure” actor for open AI—supporting the scientific lifecycle (discovery, evaluation, and deployment) rather than only publishing models. Its participation in NSF/NVIDIA-funded open infrastructure (NSF OMAI) and its expanded integration with Hugging Face for hosting and distribution are examples of this infrastructure-forward posture. As of September 6, 2026, Ai2 is led by Interim CEO Peter Clark following Ali Farhadi’s stepping down as CEO (announced in 2026 and discussed by Ai2 in its May 1, 2026 CEO interview).
At an Ai2 event marking our expanded collaboration with Providence Swedish, researchers explored the hardest problems in AI-assisted science: keeping systems steerable, grounded in human judgment and sound methods, and responsive to new evidence and experiments.
Ai2 and Providence Swedish Cancer Institute partner to advance AI-assisted scientific discoveryAi2 and Providence Swedish Cancer Institute are expanding their collaboration after AutoDiscovery helped researchers uncover and validate a promising new immune signal in invasive lobular breast cancer.
How researchers adapted Dolma for better Thai language modelsThai researchers adapted Ai2’s open Dolma toolkit to build Mangosteen, a 47-billion-token Thai corpus that filters low-quality web data while maintaining or improving model performance and strengthening Thai cultural knowledge.
How a Georgia Tech team used the open Olmo stack to trace social reasoningA Georgia Tech team used Ai2’s fully open Olmo stack to trace social reasoning back to the training data that shaped it, finding that dialogue-rich, interpersonal writing had an outsized influence on the capability.
When a model reads a drug's class from its name—not its knowledgeResearchers used Olmo 3 and its open training data to show that models can infer a drug’s class from its name instead of knowing the specific medication, and traced that shortcut to how often drugs appeared in training.
TutorMoments: Do AI tutors know when to help and when to hold back?TutorMoments is an open, replay-based evaluation framework that tests whether AI tutors can recognize when to support a student and when to hold back and encourage deeper reasoning.
Ai2 expands collaboration with Hugging Face to accelerate open scienceAi2 is expanding its partnership with Hugging Face to give its growing portfolio of fully open models, datasets, benchmarks, and applications the storage, bandwidth, and integrations needed to reach more researchers and developers.
Tracing distinctive language in AI-written textStony Brook researchers used our infini-gram engine to trace distinctive phrases in AI-generated writing back to existing sources, finding that top-selling self-published books on Amazon with substantial detected AI text overlap more heavily with rare language from previously pub
The OlmoEarth Platform: Geospatial inference at planetary scaleHow we built the OlmoEarth Platform to fine-tune geospatial models and run continent-scale satellite inference while managing massive data pipelines, distributed compute, and automatically recovering from failures at scale.
Who gets to understand AI?Why fully open models and research artifacts are essential to independent scrutiny, broader participation, and continued U.S. scientific leadership in AI.
What building Shippy taught us about building agentsBuilding Shippy taught us that reliable agents depend less on the model itself than on deterministic tools, explicit guardrails, isolated infrastructure, and evaluations grounded in real-world workflows and live data.
MolmoAct 2 shows what open models can unlock for roboticsRobotics engineer Binh Pham used MolmoAct 2 to build a voice-controlled robot that won South Park Commons’ embodied AI hackathon.
Ai2 published a recap-style article describing how scientists can steer AI-assisted discovery during research projects, referencing an August 27 event connected to the Providence partnership.
Ai2 and Providence Swedish Cancer Institute partner to advance AI-assisted scientific discoveryAi2 announced a partnership with the Paul G. Allen Research Center (PARC) at Providence Swedish Cancer Institute to apply AutoDiscovery across cancer research datasets, including a locally deployed setup aimed at keeping clinical data secure.
Ai2 expands collaboration with Hugging Face to accelerate open scienceAi2 announced an expanded Hugging Face partnership focused on storage/bandwidth and distribution for Ai2’s growing portfolio of open models, datasets, and benchmarks; it also highlights specific Hub/benchmark integrations and traffic growth claims.
Open by design: Ai2 brings fully open AI infrastructure online with NSF OMAIAi2 announced bringing NSF OMAI compute infrastructure online, framed as making open model development and ecosystem reuse possible at operational scale.
What’s next for Ai2: A conversation with Interim CEO Peter ClarkAi2 published an interview with Interim CEO Peter Clark emphasizing Ai2’s open-science mission, long-term focus, and ongoing work areas including AI for science (Asta/AutoDiscovery) and AI for the planet (OlmoEarth).
A decade of real-time intelligence for the planetAi2 marked Earth Day 2026 with an impact overview for EarthRanger and Skylight, including program scale claims about coverage and operations.