Who's who · Open-source projects
EleutherAI
Non-profit research institute building open AI models, datasets and evaluation tools
At a glance
- Founded
- July 2020
- Based in
- Online (distributed team)
- People
- Stella Biderman, Executive director; Nora Belrose, Head of interpretability; Quentin Anthony, Head of HPC; Aviya Skowron, Head of policy and ethics
- Known for
- The Pile, GPT-NeoX-20B, Pythia, lm-evaluation-harness
- Licence
- Varies by project (lm-evaluation-harness: MIT)
- Funding
- Non-profit; funded by grants and sponsors
- Staff
- 15 full- and part-time
- Model downloads
- 100M+
- Website
- https://www.eleuther.ai
EleutherAI is a non-profit research institute that builds open-source AI models, datasets and tools, and studies how models learn, how to evaluate them and how to make open-weight systems safer. Its work falls into three areas: open AI infrastructure, scientific research on training and evaluation, and safety research, including interpretability, the study of what happens inside a model.
It began in July 2020 as a Discord server where Connor Leahy, Sid Black and Leo Gao gathered volunteers to replicate GPT-3 in the open; none of the three is on the current staff list. The group released The Pile dataset in 2020, the GPT-Neo and GPT-J models in 2021, GPT-NeoX-20B in early 2022 and the Pythia model suite in 2023. Its lm-evaluation-harness is a widely used tool for testing language models, and its models have been downloaded more than 100 million times.
EleutherAI incorporated as a non-profit in early 2023 and is led by executive director Stella Biderman, with 15 full- and part-time staff plus volunteers. Its sponsors include Google's TPU Research Cloud, Open Philanthropy, Mozilla and Omidyar Network. In June 2025 it released the Common Pile v0.1, an 8-terabyte dataset of openly licensed and public domain text, with two 7-billion-parameter Comma models trained on it.
Sources
- EleutherAI, About EleutherAI, 2026-10-02
- EleutherAI, Staff, 2026-10-02
- TechCrunch, EleutherAI releases massive AI training dataset of licensed and open domain text, 2025-06-06
- GitHub, EleutherAI/lm-evaluation-harness repository, 2026-10-02
Facts last checked 2026-10-02