Zoo Labs Foundation
45+ open-source frontier models from 600M to 480B parameters. Open weights, community-trained, formally verified. Built on Qwen3+ architecture.
45+
Open Models
1T+
Max Parameters
59.2%
SWE-bench (Code)
82.7%
AIME24 Accuracy
General purpose language models for text generation, reasoning, and instruction following.
zen4-nano
zen4-mini
zen4-flash
zen4-pro
zen4-max
zen4-ultra
Specialized for software engineering. 59.2% SWE-bench verified.
zen4-coder-flash
zen4-coder-pro
Vision, audio, video, and 3D generation from a single architecture.
zen-omni
zen-vl
zen-3d
zen-director
Content safety, alignment verification, and guardrails for production deployment.
zen-guard
zen-guard-pro
zen-guard-vision
Open-source AI training platform. 100+ model support, multi-GPU distributed training, 8 training methods. 2.5 years of community development.
Training-Free GRPO
99.8% cost reduction. $18 vs $10,000+ for traditional fine-tuning.
LoRA / QLoRA
Parameter-efficient fine-tuning for single-GPU training.
DPO
Direct preference optimization without reward model overhead.
Full Fine-Tuning
Multi-GPU distributed training with DeepSpeed ZeRO-3.
DSO Protocol
Decentralized semantic optimization via experience ledger.
RLHF
Reinforcement learning from human feedback for alignment.
Run Zen models locally. Free, open-source, private by default. Supports all Zen model sizes that fit your hardware. Automatic model download from HuggingFace.
Local Inference
Run models on your GPU. No data leaves your machine.
Multi-Platform
macOS, Windows, Linux. Built with Tauri.
Mine $AI
Contribute GPU compute to Zoo Network and earn $AI tokens.
Access Zen models via the Zoo API at api.zoo.network
# Chat completions
POST /v1/chat/completions
# Embeddings
POST /v1/embeddings
# Models list
GET /v1/models
Every donation funds open-source models, research papers, and formally verified systems. 501(c)(3) tax-deductible.