0zhr is an open conversational language model designed for reasoning, long-form dialogue, and tool-aware agent workflows.
Unlike lightweight chat models optimized only for response speed, 0zhr is designed to spend additional computation on planning, structured reasoning, and maintaining coherent conversations across longer interactions.
Source code & release:github.com/0penAGI/0pen — dataset pipeline, LoRA training script (lora.py), adapter weights, and GGUF conversion.
Highlights
🧠 Thinking / reasoning capable
🔧 Designed for tool calling workflows
💬 Strong conversational abilities
🌍 Multilingual (English + Russian)
📚 Long-form explanations
🤝 Natural dialogue style
⚡ Optimized for local inference
Philosophy
0zhr was trained with a focus on dialogue quality rather than benchmark optimization alone.
The goal is to produce responses that remain coherent during long conversations, reason before answering when appropriate, and integrate naturally into agent systems.
Capabilities
reasoning over multi-step problems
coding assistance
mathematics
writing
summarization
brainstorming
analysis
roleplay
tool-use planning
instruction following
Intended Use
0zhr is suitable for
local assistants
autonomous agents
research
coding
educational applications
creative writing
chatbot systems
Comparison
Model
Speed
Thinking
Tool Calls
0pen
⭐⭐⭐⭐⭐
❌
❌
0zhr
⭐⭐⭐⭐
✅
✅
0pen prioritizes latency.
0zhr prioritizes reasoning quality.
Training
0zhr is a LoRA fine-tune of Qwen3 4B (mlx-community/Qwen3-4B-4bit), trained on a dialogue-focused corpus emphasizing
coherent conversation
reasoning
reflective dialogue
instruction following
helpfulness
multilingual communication
The full training script (lora.py) is included in this repository. This release is the GGUF (Q4_K_M) export, fully compatible with llama.cpp and Ollama.
Example
User
Why does this mathematical proof fail?
0zhr
The proof becomes invalid when it divides by (a - b). Since the assumption is a = b, we have (a - b) = 0, making that step a division by zero. Every step after that no longer follows logically.
Run with Ollama
bash
1ollama create 0zhr -f Modelfile
2ollama run 0zhr
Modelfile:
FROM ./0zhr_q4.gguf
SYSTEM """
You are 0zhr, created by 0penAGI.
"""
PARAMETER temperature 0.55
PARAMETER top_p 0.9
PARAMETER num_ctx 32768
Run with llama.cpp
llama-cli -m 0zhr_q4.gguf -p "Hello, what can you do?" -n 256
