The Zen AI Model Family

Zen fine-tunes the best open-weight model of each era. Across this family that is the Qwen3 line from Alibaba Cloud — base, VL, Omni, Embedding, Reranker, TTS, ASR, Guard, Coder — with media models on permissive bases (Wan, FLUX, TRELLIS, YuE). Hanzo adds only identity training, agentic-data fine-tuning, and abliteration. There is no from-scratch Zen model.

Base models & attribution

Language / VL / Omni / Coder / Guard / Embedding / Reranker / TTS / ASR: fine-tuned from Qwen3 (Alibaba Cloud, Apache-2.0). The reference language checkpoint fine-tunes Qwen/Qwen3-8B.
Media (image / video / 3D / audio): permissive bases — FLUX.1-schnell (Apache-2.0), Wan2.2, TRELLIS (MIT), YuE.

Each model card on huggingface.co/zenlm lists its exact base_model and license.

Documentation

Quick start

pip install transformers torch accelerate

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("zenlm/zen-eco-4b-instruct")
tokenizer = AutoTokenizer.from_pretrained("zenlm/zen-eco-4b-instruct")

Citation

@article{zen2025,
  title  = {The Zen AI Model Family},
  author = {Hanzo AI and Zoo Labs},
  year   = {2025}
}

License

Apache-2.0 — see LICENSE and NOTICE. Upstream Qwen3 is Apache-2.0.

Built by Hanzo AI and Zoo Labs Foundation

1.8 KiB Raw Permalink Blame History