ACPACP AI Research
⌕ZH
ResourceHeat 85Quality 90

Cohere North Mini Code Open-Source Agentic Coding Model

Cohere has released North Mini Code, an open-source agentic coding model designed for agentic software engineering tasks. It features a Mixture-of-Experts architecture for efficient inference on a single Nvidia H100 GPU and supports a large context window.

AIOpen SourceCoding ModelAgentic AILLMCohereHugging Face

Resource Overview

Cohere introduced North Mini Code in June 2026, an innovative open-source agentic coding model specifically engineered for agentic software engineering tasks [1]. This model aims to provide an efficient and accessible solution for developers tackling complex coding challenges. It signifies a strategic move by Cohere to offer open-source alternatives in the AI landscape, addressing concerns around transparency, security, and vendor dependence in enterprise AI applications [2, 3].

Key Content

North Mini Code is a 30-billion-parameter Mixture-of-Experts (MoE) model, uniquely designed with only 3 billion parameters active per token, which contributes to its inference efficiency [1]. The model is capable of advanced functionalities such as sub-agent orchestration, architecture mapping, code review, and terminal work [1]. It boasts a substantial context window of 256,000 tokens and a maximum generation length of 64,000 tokens [1].

Available on Hugging Face under an Apache 2.0 license, North Mini Code permits permissive commercial use [1]. A significant advantage is its ability to run on a single Nvidia H100 GPU, making self-hosting a practical option for many developers [1]. Cohere claims that North Mini Code outperforms comparable open-weight models like Alibaba's Qwen3 and Google's Gemma 4 on the Artificial Analysis Coding Index, achieving a score of 33.4 [1]. Furthermore, it reportedly delivers up to 2.8 times higher output throughput than Mistral's Devstral Small 2 on identical hardware [1]. The model was trained using a combination of supervised fine-tuning and reinforcement learning across more than 70,000 verifiable tasks derived from approximately 5,000 repositories [1].

How to Use

Developers can access and utilize North Mini Code via the Hugging Face platform [1]. Its capability to run on a single Nvidia H100 GPU makes self-hosting a practical and accessible option for developers who prefer to manage their own infrastructure [1].

Notes and Caveats

  • Independent testing indicated that North Mini Code generated three times the output tokens of comparable models, which could lead to higher verbosity costs in high-volume production workloads [1].
  • Cohere's performance claims against other models are based on their internal or commissioned benchmarks (Artificial Analysis Coding Index), and independent verification may be beneficial to confirm these findings [1].
  • While open-source, developers who prefer not to manage their own infrastructure would need to access it via an API, which might incur costs [1].

Sources

  • [1] VentureBeat, "Cohere open-sources a coding agent that runs on a single H100", 2026-06-09
  • [2] The New Stack, "Cohere sold sovereign AI to enterprises, now it's targeting developers with its first coding model", 2026-06-15
  • [3] BetaKit, "Cohere releases its most powerful AI model as open source", 2026-05-22