SuperAPI AI Directory
Partner with us中文AI newsFavorites
DirectoryNews and tutorials

North Mini Code - Cohere's Open-Source Agentic Coding Model

Author: AI小集 · Original source: AI工具集

What is North Mini Code

North Mini Code is an open-source agentic coding model released by Cohere . The model adopts a MoE architecture with 30B total parameters and only 3B activated, supporting 256K context and 64K output. The model is optimized for code generation and software engineering tasks, scoring 33.4 on the Artificial Analysis Coding Index, with an API output speed of approximately 199 tokens/second, and can be deployed locally on a single H100 at minimum.

North Mini Code

Main Features of North Mini Code

  • Code Generation: The model supports complex code generation tasks involving scientific computing and algorithmic reasoning.
  • Agentic Software Engineering: Can perform repository-level code modifications in frameworks such as SWE-Agent and OpenCode.
  • Terminal Agent: Drives Shell tools end-to-end to complete multi-turn build and deployment tasks.
  • Local Coding: 3B activated parameters support low-latency local inference, suitable for privacy-sensitive scenarios.
  • Sub-agent Orchestration: Understands system architecture and coordinates sub-agents to complete code review and refactoring.

Technical Principles of North Mini Code

  • MoE Sparse Activation: Only 3B of the model's 30B total parameters are routed and activated, significantly reducing inference cost and VRAM usage.
  • Ultra-long Context: The 256K input context supports loading medium-sized codebases at once, while 64K output can generate complete modules.
  • Agentic Training: Trained for multiple frameworks including SWE-Agent and ReAct terminal toolchains, generalizing across different Agent scaffolds.
  • Speed Optimization: Under the same concurrency and hardware, output throughput is 2.8 times higher than Devstral Small 2, with 30% lower inter-token latency.

How to Use North Mini Code

  • Download Weights from Hugging Face: Obtain the open-source weights and deploy locally via compatible frameworks.
  • Cohere API Call: Use the Chat V2 API to try it at zero cost within the free quota.
  • Model Vault Hosting: Enterprise users can deploy to production environments with one click on the Cohere hosted inference platform.
  • OpenCode Integration: On the OpenCode platform, directly select the north-mini-code-free model for a free trial.

Core Advantages of North Mini Code

  • Efficient and Low-Cost: The model's 3B activated parameters make inference cost far lower than dense models of comparable performance.
  • Enterprise Compliance-Friendly: Supports local, VPC, or private cloud deployment, meeting data residency requirements in industries such as finance.
  • Long Window Advantage: The 256K/64K context and output length are suitable for large codebase analysis and batch generation.
  • Multi-Framework Compatibility: Natively adapts to OpenCode while also compatible with mainstream Coding Agent scaffolds.

North Mini Code Project Address

  • Project Official Website：https://docs.cohere.com/docs/north-mini-code-1.0
  • HuggingFace Model Repository：https://huggingface.co/blog/CohereLabs/introducing-north-mini-code

Comparison of North Mini Code with Similar Competitors

Dimension North Mini Code Devstral Small 2
Architecture MoE（30B / 3B） Dense Model
Output Throughput 2.8 times higher Baseline
Inter-Token Latency 30% lower Baseline
Maximum context 256K Shorter
Maximum output 64K Shorter
Open-source license Apache 2.0 Depends on version
Positioning Agentic coding + terminal tasks General-purpose coding assistance

Application scenarios of North Mini Code

  • Enterprise private code assistant: The model supports local or VPC deployment, meeting the compliance requirements of industries such as finance and healthcare that core code must not leave the domain.
  • Repository-level refactoring: The 256K context window can load a medium-sized codebase at once, performing cross-file dependency analysis and batch refactoring.
  • Terminal automation pipeline: Drives the Shell toolchain through frameworks such as SWE-Agent to complete automated build, test, and deployment tasks.
  • Scientific computing assistance: Generates algorithm implementations, data processing scripts, and complex mathematical reasoning code for researchers.
  • IDE real-time plugin: 3B activated parameters support low-latency inference, making it suitable for embedding into editors to provide real-time code completion and intelligent suggestions.