Particle.news

Cohere Releases North Mini Code as Open, Self-Hostable Coding Model

The Apache 2.0‑licensed Mixture‑of‑Experts model is designed to run on a single H100 GPU to give developers direct control over agentic coding tools.

Overview

  • Cohere launched North Mini Code, its first coding model, which launched on Monday as an open‑weight model available under an Apache 2.0 license.
  • The model is a 30 billion‑parameter Mixture‑of‑Experts design with about 3 billion active parameters so it can run on a single Nvidia H100 GPU for practical self‑hosting.
  • Cohere offers the weights on Hugging Face and API access through its platform and says internal benchmarks show strong code and agentic task performance and higher throughput versus some peers.
  • Industry analysts and reporting place the release in a wider shift toward permissively licensed, developer‑focused models that let firms avoid high API bills and vendor control over access.
  • For developers and regulated organizations this means more control over where models run and how they are operated, which could cut costs, reduce latency, and guard against vendor or regulatory interruptions.