Meet ‘North Mini Code’: Cohere’s 30B Open-Weight Mixture-of-Experts Model With 3B Active Parameters for Agentic Coding

Cohere has unveiled North Mini Code, a significant addition to its open-source AI model portfolio designed specifically for coding applications. This 30-billion parameter mixture-of-experts model represents the company's first dedicated model for developer-focused coding tasks, offering an efficient alternative to larger language models while maintaining competitive performance for programming-related workloads.

North Mini Code is a mixture-of-experts architecture containing 30 billion total parameters, but only 3 billion parameters are actively used during inference. This design allows the model to operate efficiently on a single H100 GPU—a crucial advantage for organizations with limited computational resources. The model supports an impressive 256K context length, enabling developers to work with lengthy codebases and comprehensive documentation within single requests. This combination of efficiency and capability positions North Mini Code as a practical solution for enterprises seeking cost-effective AI-powered development tools without sacrificing functionality.

The release of North Mini Code has several important ramifications for the AI and software development sectors:

• Democratized Access: Open-weight models reduce barriers to entry for developers and organizations lacking access to proprietary AI systems or significant computational resources
• Cost Efficiency: The mixture-of-experts architecture with limited active parameters reduces operational expenses while maintaining performance quality
• Agentic Capabilities: The model is specifically optimized for autonomous coding tasks, supporting the growing trend of AI agents handling development workflows
• Extended Context Window: The 256K context length enables processing of large codebases, documentation, and related materials simultaneously
• Infrastructure Flexibility: Operating on a single H100 makes deployment more accessible than models requiring multiple GPUs or specialized hardware

North Mini Code represents a strategic move by Cohere to compete in the specialized coding AI segment, challenging established players like OpenAI and Anthropic. By releasing an open-weight model with efficient architecture, Cohere enables broader adoption of AI-assisted development while maintaining technical sophistication. The focus on agentic coding—where AI systems autonomously write and modify code—aligns with industry trends toward automation and increased developer productivity. This release underscores the shift toward specialized, efficient AI models tailored for specific professional domains rather than general-purpose systems.