Open AI Research for Everyone
Zoo Labs Foundation is a 501(c)(3) nonprofit advancing open-source AI research, frontier model development, and decentralized science.
Our Mission
Advance open AI research by building frontier models, publishing research papers, and developing tools that make AI accessible to researchers worldwide.
45+
Open Models
Frontier models from 600M to 480B parameters, all open weights
7
Peer-Reviewed Papers
Published research in HLLM, DSO, and formal verification
102
ZIP Specifications
Community-governed improvement proposals
Our Story
Founded in 2021, Zoo Labs Foundation started with the belief that frontier AI should be open, verifiable, and community-governed. Not locked behind corporate walls.
We built the Zen model family on Qwen3+ architecture: 45+ models spanning text, code, vision, audio, video, and 3D generation. Every model has open weights on HuggingFace.
Our research spans Hamiltonian Large Language Models (HLLMs), decentralized semantic optimization (DSO), and 15 formally verified proofs in Lean 4. We publish everything open access.
Zoo Gym, our training platform, has enabled thousands of researchers to train and fine-tune models at 99.8% lower cost through Training-Free GRPO.
Core Values
Open by Default
Open weights, open papers, open code. Closed AI is a dead end for science.
Formally Verified
15 Lean 4 proofs. Every critical system machine-checked, not hand-waved.
Community Governed
102 ZIPs. Research direction decided by contributors, not executives.
Safe by Design
Zen-guard safety models. Alignment research integrated, not bolted on.
Leadership
Antje Worring
Executive Director & Co-Founder
Building safer AI for children since founding Zoo Labs Foundation in 2021.
What We Do
Frontier Model Research
Developing the Zen family of open-source models. From 600M parameter edge models to 480B parameter frontier systems, all with open weights on HuggingFace.
zen4-ultra 480B
Training Infrastructure
Zoo Gym: open-source platform supporting 100+ models, 8 training methods, and multi-GPU distributed training. 99.8% cost reduction via Training-Free GRPO.
Zoo Gym
Decentralized AI Network
Zoo Network: decentralized AI compute with recursive self-learning. Post-quantum secure consensus. FHE for private inference.
Zoo Network
Ecosystem
Built on the Lux-Hanzo-Zoo stack. Post-quantum consensus, AI compute chain, and open research network.
Lux Network
Post-quantum L0 consensus with sub-second finality.
Hanzo AI
L1 AI compute chain. Mining, model serving, and infrastructure.
Zoo Network
Application layer for decentralized AI research and governance.
Advance Open AI Research
Fund open-source models, contribute to ZIPs, or train models on Zoo Gym. Every contribution is tax-deductible.