jan-v1-4B-npu

Jan-v1-4B

Model Description

Jan-v1-4B is a 4-billion-parameter open-weight model developed by JanHQ.
It is based on the Qwen3-4B-Thinking architecture and optimized for agentic reasoning, planning, and tool use, while remaining efficient for local deployment.

Trained on a curated dataset for reasoning and conversation, Jan-v1-4B balances factual accuracy, dialogue quality, and usability in compact form.

Features

• Agentic reasoning: decomposes tasks and plans multi-step workflows.
• Tool integration: supports structured tool calling and external function use.
• Conversational fluency: strong dialogue and instruction-following.
• Efficient deployment: available in quantized GGUF formats for local devices.
• High factual accuracy: achieves 91.1% on SimpleQA benchmark.

Use Cases

• Automation agents and reasoning workflows
• Virtual assistants with tool calling
• Research and tutoring support
• Local deployment on consumer hardware
• Fine-tuned domain-specific applications

Inputs and Outputs

Input:

• Text prompts or conversation history
• Structured reasoning tasks with tool calls

Output:

• Generated text (answers, plans, explanations)
• Structured tool-call responses

---

How to use

⚠️ Hardware requirement: the model currently runs only on Qualcomm NPUs (e.g., Snapdragon-powered AIPC).
Apple NPU support is planned next.

1) Install Nexa-SDK

• Download and follow the steps under "Deploy Section" Nexa's model page: Download Windows arm64 SDK
• (Other platforms coming soon)

2) Get an access token

Create a token in the Model Hub, then log in:

nexa config set license '<access_token>'

3) Run the model

Running:

nexa infer NexaAI/jan-v1-4b-npu

---

License

This model is released under the Creative Commons Attribution–NonCommercial 4.0 (CC BY-NC 4.0) license.
Non-commercial use, modification, and redistribution are permitted with attribution.
For commercial licensing, please contact [email protected].

References

• Model card: https://huggingface.co/janhq/Jan-v1-4B