Cohere launches North Mini Code, its first developer-focused model, on Hugging Face

North Mini Code is Cohere's first model aimed squarely at developers, using a 30B-parameter Mixture-of-Experts design with just 3B active parameters — an efficiency play that keeps inference cheap while targeting agentic coding tasks. It ships openly under Apache 2.0 on Hugging Face, fitting the week's open-weights momentum alongside DiffusionGemma.

Cohere claims North Mini Code outperforms several larger open-source models on coding benchmarks, positioning it as a cost-efficient option for code generation and agentic software workflows. The MoE-with-small-active-params architecture mirrors a broader industry shift toward sparse models that deliver capability without dense-model inference cost — directly relevant to the 'tokenmaxxing' economics debate Nadella raised.

The launch also reflects Cohere's pivot: long focused on enterprise RAG and retrieval, it is now courting developers directly. CEO Aidan Gomez has emphasized Cohere's Canadian roots and sovereign-AI positioning, including new government partnerships in Québec. North Mini Code's competitiveness against larger open models will be tested against Qwen, DeepSeek, and Gemma variants that dominate the open-source coding landscape.