🧠 Jan-Nano 4B

A quantized version of Jan-Nano 4B, designed for deep research tasks, autonomous tool use, and integration with the Model Context Protocol (MCP). This model is optimized for use with Ollama and supports local agentic workflows.

📦 Model Overview

• Base: Qwen3 architecture
• Size: 4B parameters (quantized)
• Context Length: Up to 128K tokens
• Specialty: Deep reasoning, tool-calling via MCP, compact size
• Quantization: Q8 optimized for performance and compatibility

🚀 Quickstart with Ollama

1. Pull the Model

ollama pull yasserrmd/jan-nano-4b

2. Run the Model Locally

ollama run yasserrmd/jan-nano-4b

⚙️ Recommended Parameters

• temperature: 0.7
• top_p: 0.8
• top_k: 20
• min_p: 0.0
• Optional: --hidethinking to suppress self-verbalization

Example:

ollama run yasserrmd/jan-nano-4b \
  --temperature 0.7 --top-p 0.8 --top-k 20 --hidethinking

📡 Features

✅ MCP tool-calling (search, summarize, calculator, etc.) ✅ Optimized for 128K token context length ✅ Suitable for research, question answering, and autonomous agents ✅ Fast inference on consumer GPUs (8–16GB VRAM)

📝 Example Usage

Prompt:

User: /no_think
How has AI improved weather forecasting in the last five years?

Model Output:

• Automatically calls search tools via MCP
• Synthesizes research into clear summaries
• Provides references and statistics

🧩 Integrations

• Can be used with external orchestration layers like LangGraph, Ollmcp, or custom HTTP MCP servers
• Works well with tool templates and system prompts

📁 License & Source

• Model: Menlo/Jan-nano on Hugging Face