Research institute · Research orgs · Seattle, Washington · Est. 2014
Allen Institute for AI (Ai2) is a Seattle-based nonprofit AI research institute founded in 2014 by the late Paul Allen, focused on building and advancing “truly open” AI to solve major scientific and societal problems. Ai2’s stated mission is to build breakthrough AI to solve the world’s biggest problems, emphasizing transparency, reproducibility, and broad community access to research artifacts (data, code, model weights, training recipes, and evaluation tools).
Strategically, Ai2 operates like a public research lab with multiple model and application lines: (1) open language and multimodal model development (e.g., Olmo and Molmo families), (2) open post-training and training infrastructure aimed at making model behavior understandable and extensible (e.g., Tülu 3 and Olmo 3’s “model flow” concept), and (3) AI systems for applied scientific discovery (e.g., AutoDiscovery and related evaluation/agent ecosystems). Institutionally, Ai2 also acts as a compute-and-infrastructure steward for open scientific AI, including an NSF/NVIDIA-funded effort (NSF OMAI) to build a national open multimodal AI infrastructure and develop fully open AI models for the U.S. scientific community.
No curated programs yet. Check the careers page ↗.
LitPivot helps researchers refine early research ideas through literature-supported critiques and revisions as their proposals evolve.
OlmoEarth v1.3: State-of-the-art embeddings for the whole planetOlmoEarth v1.3 brings state-of-the-art satellite embeddings, precomputed global data, and a browser-based Embeddings Explorer for mapping and analyzing the planet.
FairStress: A deeper evaluation of fairness in LLM decision-makingFairStress is a new benchmark that reveals demographic differences in how LLMs defend and explain their decisions, even when their final choices appear fair.
HiRO-ACEv2: Simulating extreme weather at 3 km resolution for a fraction of the costHiRO-ACEv2 captures extreme rain, wind, and pressure events at 3 km resolution for about 1/300th the cost of X-SHiELD, making detailed global climate simulations more accessible to researchers.
How Molmo 2’s open release helped an MIT researcher rethink fish trackingMolmo 2’s open release helped MIT researcher Kai Van Brunt develop Molmo2Fish, an experimental model that tracks salmon in sonar footage and corrects some tracking errors using everyday language.
Ai2 at COLM 2026: Open research, from models to agents for scienceAi2 joined COLM 2026 to share advances in open language models, training infrastructure, and AI for science, including Olmo Hybrid, Asta, and Bolmo, while connecting with the broader research community.
Now in Nature: Retrofitting language models to operate over bytesThe technique behind Bolmo, Ai2’s fully open byte-level language models, is now published in Nature, with new checkpoints showing the approach generalizes beyond Olmo to other model families.
Open-sourcing AstaBrief, the fast report-generation model in AstaWe’re releasing AstaBrief, an 8B open-weights model for generating cited scientific reports, available in Asta’s Fast mode or to download and run on your own infrastructure.
Introducing Olmo-core 3: Open, scalable training infrastructure for large MoEsOlmo-core 3 introduces a redesigned, fully open training stack for efficiently scaling mixture-of-experts models into the trillion-parameter range.
What a crowdsourced game revealed about steering Olmo 3A crowdsourced game built on Olmo 3 showed how people can exploit unexpected model behaviors to stress-test prosocial AI evaluations—and how open access to a model’s internals can help researchers understand why those tests break.
Teaching future scientists to interrogate AI tools for scientific discoveryUniversity of Washington students put Ai2’s AutoDiscovery to the test, showing how AI can surface promising scientific leads while making human judgment, domain expertise, and rigorous validation more important than ever.
How Goodfire used Ai2’s open post-training stack to trace unwanted model behaviorGoodfire used Ai2’s fully open post-training stack to predict LLM behavioral changes, trace unwanted model behavior back to individual training examples, and test targeted fixes without sacrificing broader capability gains.
Ai2 released MolmoAct 2 and reported open code/data releases, including integration into Hugging Face’s LeRobot platform; Ai2 also described training-data releases and evaluation outcomes.
What’s next for Ai2: A conversation with Interim CEO Peter ClarkAi2 published a CEO interview describing Ai2’s focus on long-term, high-impact research and emphasizing the institute’s openness commitment amid rapid progress in AI.
MolmoWeb: An open agent for automating web tasksAi2 released MolmoWeb with a published codebase (training code, evaluation harness, annotation tool, synthetic data generation pipeline, and demo client code) and described agent evaluation targets.
MolmoBot: Training robot manipulation entirely in simulationAi2 described and released an expanded MolmoBot stack, including updates such as a code/data release and a dataset of expert manipulation trajectories across objects and procedurally generated environments.
Olmo 3: Charting a path through the model flow to lead open-source AIAi2 announced Olmo 3 and framed it around a fully documented model flow exposing stages, datasets, and training components, including multiple Olmo 3 pathways and checkpoints.
NSF and NVIDIA partnership enables Ai2 to develop fully open AI models to fuel U.S. scientific innovationNSF reported the combined $152M NSF+NVIDIA support for Ai2’s NSF OMAI effort to build national open multimodal AI infrastructure and develop fully open AI models for scientific use.