Allen Institute for AI
Ai2 is a Seattle-based nonprofit AI research institute that builds breakthrough AI to solve the world's biggest problems. It develops foundational research and innovation—including fully open models, data, robotics, conservation and scientific AI—for public benefit.
At a Glance
- Academic and independent AI researchers
- AI developers and open-source communities
- Scientists and research institutions
- Government and public-sector science programs
- +3 more
AI Tools by Allen Institute for AI
(7)
SUPER Benchmark
LLM Agent Research Repo Benchmark
MolmoWeb
Open Multimodal Web Agent
olmOCR
Open Source PDF OCR Toolkit
AutoDiscovery
AI Dataset Exploration Tool
ASTA
AI Scientific Literature Synthesis
SciArena
Scientific Literature Model Arena
OLMo
Open Source LLM for Research
Discussions
No discussions yet
Be the first to start a discussion about Allen Institute for AI
Latest News
OlmoEarth v1.3: State-of-the-art embeddings for the whole planet
Ai2 at COLM 2026: open research from models to agents for science
Now in Nature: Ai2's Bolmo work on retrofitting language models to operate over bytes
Open-sourcing AstaBrief, an 8B fast report-generation model for cited scientific reports
Products & Services
Fully open language models and a complete model flow, including open data, training code, checkpoints, evaluation and documentation; current Olmo 3 variants include 7B and 32B Base, Think and Instruct models.
Open multimodal AI models for image and video understanding, visual grounding/pointing, counting, captioning and reasoning across documents and images.
Open multimodal/video model family with 4B, 8B and fully open end-to-end O-7B variants for video understanding, captioning, pointing, tracking and counting.
Agentic ecosystem for scientific discovery, including research assistants, AstaBench evaluation benchmarks and tools that help scientists frame questions, trace evidence and identify open issues.
Market Position
Ai2 differentiates itself as a public-interest, nonprofit alternative to closed commercial AI labs by publishing genuinely open model flows—data, code, weights, evaluations and documentation—while pursuing frontier performance. Its closest competitive set varies by project and includes Meta's Llama, Mistral, Qwen and other open-weight model developers; Hugging Face and academic labs in open-model infrastructure; and commercial frontier labs such as OpenAI, Anthropic and Google for model capability. In scientific agents, literature discovery and Earth observation, it competes with specialized research platforms and geospatial AI providers.
Leadership
Founders
Paul Allen
Microsoft co-founder, philanthropist and technology investor; he founded Ai2 in 2014 to find transformative ways to develop AI for major global challenges.
Executive Team
Peter Clark
Interim Chief Executive Officer; Senior Research Director
Founding member of Ai2 and co-leader of the Asta project; a researcher in natural-language understanding and scientific AI.
Ali Farhadi
Former Chief Executive Officer (July 2023–March 2026)
AI researcher and professor at the University of Washington; previously worked in AI/ML at Apple and later joined Microsoft's AI organization.
Board of Directors
Founding Story
Paul Allen established Ai2 in 2014 with the vision of creating a research institute that could pursue transformative artificial-intelligence research and engineering in service of the common good. The institute's current emphasis on open science and open models extends that original public-interest vision.
Business Model
Revenue Model
Ai2 is a nonprofit research institute funded through philanthropy, grants and research partnerships. Its core models, datasets, code and benchmarks are generally released openly; the public product pages emphasize free downloads, documentation and research access rather than subscription pricing.
Target Markets
- Academic and independent AI researchers
- AI developers and open-source communities
- Scientists and research institutions
- Government and public-sector science programs
- Environmental, conservation and climate organizations
- Robotics researchers and developers
- AI research, reproducibility and model interpretability
- Building and fine-tuning language and multimodal models
- Scientific literature review, hypothesis generation and AI-assisted discovery
- Video understanding, visual QA, image grounding and robotics
- Earth observation, crop mapping, wildfire risk, mangrove monitoring and ocean intelligence
- Benchmarking and evaluating autonomous LLM agents
- University of Washington Paul G. Allen School of Computer Science & Engineering
- Providence Swedish Cancer Institute / Paul G. Allen Research Center
- Wetlands International and Global Mangrove Watch
- NASA Jet Propulsion Laboratory