North Mini Code marks the debut release of Cohere's North family and represents the company's first agentic coding model. It is built as a sparse mixture-of-experts architecture with 30 billion total parameters but only 3 billion active per inference, an efficiency-oriented design that targets high performance while keeping compute requirements modest. Cohere developed and optimized it specifically for software engineering tasks, framing it as a foundation for handling complex developer workflows and large codebases in agentic settings.

With its expansive context window, North Mini Code is well suited to multi-file refactors, repository-scale reasoning, and long-running agentic coding sessions that need to keep large amounts of source in view at once. The sparse-MoE design suggests a balance between capability and inference cost, making it attractive for coding assistants that must sustain many turns without ballooning latency or spend. It is positioned as a practical fit for developers looking for a Cohere-built coding model that can power autonomous tooling pipelines and interactive IDE-style assistants.