Jan-v1-4B: Agentic Language Model

Jan-v1-4B is the inaugural release in the Jan Family of models, developed by janhq and built upon the Qwen3-4B-thinking architecture. This 4 billion parameter model is specifically engineered for agentic reasoning and problem-solving, featuring a substantial 40960 token context length. It aims to provide enhanced capabilities for complex tasks and tool utilization, particularly within the Jan App.

Key Capabilities

• Agentic Reasoning: Designed for advanced problem-solving and tool integration.
• Enhanced Question Answering: Achieves 91.1% accuracy on SimpleQA, demonstrating strong factual recall and understanding.
• Optimized for Jan App: Seamlessly integrates with the Jan App for immediate access to its features.
• Conversational Performance: Evaluated for robust conversational and instructional capabilities through chat benchmarks.

Good For

• Agentic Workflows: Ideal for applications requiring sophisticated reasoning and tool use.
• Factual Question Answering: Suitable for tasks demanding high accuracy in retrieving and synthesizing information.
• Local Deployment: Supports local inference via vLLM and llama.cpp, offering flexibility for developers.
• Jan App Users: Provides an optimized model experience directly within the Jan App ecosystem.