Cohere has launched North Mini Code, its first open-source model for developers and the inaugural model in its next generation of AI systems. Built as a mixture-of-experts (MoE) model, it is designed for agentic coding, software engineering, and terminal tasks. The model has 30 billion total parameters, with only 3 billion active at a time, making it relatively efficient to run. It also supports a long context window of 256K tokens and a maximum generation length of 64K tokens. The model is released under an Apache 2.0 license and is intended to support Cohere’s goal of making sovereign AI practical for developers.
North Mini Code is available through Hugging Face, Cohere API, Model Vault, and OpenRouter, and can be used with tools such as OpenCode. Cohere says it can run on at least one H100 GPU at FP8. In benchmark testing, it showed competitive performance against similarly sized open-source models, scoring 33.4 on the Artificial Analysis Coding Index. The company also reports strong speed advantages, including up to 2.8x higher output throughput than Devstral Small 2 and a 30% improvement in inter-token latency.
Cohere emphasizes that the model is meant for developers who want control over their AI infrastructure, including on-premises or local deployment. It is positioned as the first of several upcoming open models, with community feedback expected to shape future releases.
Article timeline
Build this article’s timeline
Extract dated events, key actors, and open questions using the same timeline analysis available in Signal.
Comments
Login to view and post comments.