Chat & Reasoning Models

The Zen Chat & Reasoning family provides a complete spectrum of language models for conversational AI and multi-step reasoning, from ultra-lightweight edge models (0.6B) to frontier reasoning systems (1M context). Built on the Zen MoDE architecture (Mixture of Diverse/Distilled Experts), these models deliver strong performance across general chat, specialized reasoning, and domain-specific tasks.

Model Gallery

Quick Start

Local Inference with Transformers

Use any model from the table above. Here's an example with Zen Eco 4B Instruct:

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "zenlm/zen-eco-4b-instruct"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id, 
    torch_dtype="auto", 
    device_map="auto"
)

messages = [{"role": "user", "content": "Explain quantum computing in one sentence."}]
text = tokenizer.apply_chat_template(
    messages, 
    tokenize=False, 
    add_generation_prompt=True
)
inputs = tokenizer([text], return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True))

Using the Zen API

For production deployments, use the OpenAI-compatible API at api.hanzo.ai:

curl https://api.hanzo.ai/v1/chat/completions \
  -H "Authorization: Bearer $HANZO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "zen-eco-4b-instruct",
    "messages": [
      {"role": "user", "content": "Explain quantum computing in one sentence."}
    ],
    "max_tokens": 512
  }'

Get your API key at console.hanzo.ai — includes $5 free credit on signup.

Model Selection Guide

For Edge & Mobile: Choose Zen Nano (0.6B) or Zen Eco (0.75–4B) for on-device deployment with minimal latency.

For Speed & Throughput: Zen 5 Flash (4.02B, 32K) delivers sub-100ms first-token-to-finish for high-volume routing.

For Reasoning & Depth: Zen 5 Pro (512K context) or Zen 5 (1M context) for complex multi-step tasks and long-document understanding.

For Content Generation: Zen Blog and Zen Scribe (2–8B) are tuned for structured writing and article generation.

For Multilingual: Zen Multilingual (8B, 128K) covers 100+ languages with strong cross-lingual understanding.

Architecture

All Zen chat models are built on Zen MoDE, a modern Mixture of Experts architecture featuring:

• Sparse Activation: Active parameters scale independently of total capacity
• Extended Context: Up to 1M tokens for frontier models
• Grouped Query Attention (GQA): Efficient inference without sacrificing quality
• Multi-lingual: Strong performance across 100+ languages
• Apache 2.0 License: Download, fine-tune, and deploy commercially

Format Support

All models are available in multiple formats for flexible deployment:

• SafeTensors (primary) — Full precision (bfloat16) for training and fine-tuning
• GGUF (quantized) — Q4_K_M, Q5_K_M, Q8_0, F16 for CPU and edge inference
• MLX (Apple Silicon) — Metal-accelerated inference on M1/M2/M3 chips