Zen Eco is a family of 4-billion parameter language models designed for balanced performance and efficiency. Built on a 4B foundation architecture, Zen Eco delivers strong capabilities across a wide range of tasks while maintaining cost-effective deployment.

The Zen Eco family includes specialized variants:

zen-eco-4b-instruct - General instruction following
zen-eco-4b-thinking - Enhanced reasoning with chain-of-thought
zen-agent-4b - Tool calling and agent capabilities
zen-eco-instruct - Alternative instruction variant

Quick Start

from transformers import AutoModelForCausalLM, AutoTokenizer

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

# Generate text
messages = [{"role": "user", "content": "Explain quantum computing in simple terms"}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt")

outputs = model.generate(**inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Model Variants

zen-eco-4b-instruct

Best for: General-purpose instruction following and conversational AI

4B parameters
32K context window
Multiple formats: SafeTensors, GGUF, MLX
Optimized for balanced quality and speed

zen-eco-4b-thinking

Best for: Complex reasoning tasks requiring step-by-step analysis

Extended thinking capabilities with chain-of-thought
Ideal for mathematical reasoning, logical problems
18.4 GB model size (full precision)
Available in GGUF and MLX formats

zen-agent-4b

Best for: Tool calling and autonomous agent applications

Function calling support via Jinja2 templates
Agent-specific training for task planning
22.6 GB model size (full precision)
184 downloads/month (highest adoption in family)

zen-eco-instruct

Best for: Alternative instruction following approach

Efficient instruction tuning
Claims 95% less energy than cloud AI
100% local processing (privacy-focused)
Compact deployment footprint

Available Formats

All Zen Eco models support multiple deployment formats:

Format	Use Case	Precision	Compatibility
SafeTensors	Training/Fine-tuning	BF16	PyTorch, Transformers
GGUF	CPU Inference	Q4_K_M, Q5_K_M, Q8_0	llama.cpp, Ollama
MLX	Apple Silicon	4-bit	M1/M2/M3 Macs

Key Features

✅ Efficiency: 95% less energy consumption vs cloud AI
✅ Privacy: 100% local processing, no cloud required
✅ Flexibility: Multiple specialized variants for different use cases
✅ Open Source: Apache 2.0 license, fully open weights and code
✅ Multi-Format: SafeTensors, GGUF, and MLX support
✅ Context: 32,768 token context window

Use Cases

Conversational AI: Customer support, chatbots, virtual assistants
Content Generation: Writing assistance, summarization, translation
Code Understanding: Code review, documentation, explanation
Reasoning Tasks: Problem-solving, analysis, planning
Agent Applications: Tool use, task automation, workflow orchestration

Hardware Requirements

Minimum (GGUF Q4_K_M)

RAM: 8 GB
Storage: 4 GB
Platform: CPU-only (llama.cpp)

Recommended (SafeTensors BF16)

VRAM: 16 GB
RAM: 32 GB
GPU: NVIDIA RTX 3090 or better
Platform: CUDA 11.8+

Optimal (Full Precision)

VRAM: 24 GB (RTX 4090, A5000, A6000)
RAM: 64 GB
GPU: NVIDIA A100 (40GB) for best performance

Performance

While this base repository doesn't contain model weights (see variants above), the Zen Eco family delivers:

Fast Inference: 10-50 tokens/sec (depending on hardware)
Low Latency: <100ms first token (GGUF Q4)
High Throughput: Batch processing support
Energy Efficient: 95% reduction vs cloud-based inference

Training

All Zen Eco models are trained using:

Framework: zoo-gym
Base: 4B foundation architecture
Method: Supervised fine-tuning with identity training
Data: Zen agentic dataset + specialized task data

Citation

If you use Zen Eco models in your research or applications, please cite:

@software{zen_eco_2025,
  title={Zen Eco: Efficient 4B Language Model Family},
  author={Zen LM},
  organization={Hanzo AI and Zoo Labs Foundation},
  year={2025},
  url={https://huggingface.co/zenlm/zen-eco}
}

Resources

Website: https://zenlm.org
Models: https://huggingface.co/zenlm
GitHub: https://github.com/zenlm
Training Framework: zoo-gym
Documentation: Zen Docs

Model Cards

For detailed specifications and usage examples, see individual model cards:

License

All Zen Eco models are released under the Apache 2.0 License.

This means you can: ✅ Use commercially
✅ Modify and redistribute
✅ Private use
✅ Patent use

Free forever. No restrictions.

Zen AI: Clarity Through Intelligence

README.md Unescape Escape

Zen Eco - 4B Efficient Language Model Family

Model Overview