Start Here - North Mini Code

North Mini Code is Cohere’s coding-focused 30B total / 3B active parameter Mixture-of-Experts model for agentic software engineering, terminal tasks, and code generation.

Quickstart

Run a first prompt and use the recommended generation settings.

Deployment Paths

Pick between Cohere-hosted, Transformers, vLLM, SGLang, Docker, or AWS paths.

Agentic Coding Guide

Set up tool calls, chat templates, and multi-step coding-agent loops.

Model Details

Review architecture, context length, training approach, and tool-use behavior.

Core facts: model ID north-mini-code-1-0, 256K context window, 64K max output, Apache 2.0 license.

Fastest managed path
Fastest local path
Best agent path

Use Cohere’s hosted platform if you want the model without managing weights or inference servers.

North Mini Code is available on Cohere’s Chat endpoints.
Cohere’s docs position Model Vault as the production path.
Start here if you care more about speed-to-first-request than local control.

Open the Cohere deployment page

Choose your runtime

Pick the path that matches your goal:

evaluate locally: /deployment/transformers
build an agent server: /deployment/vllm or /deployment/sglang
deploy on AWS: /deployment/sagemaker or /deployment/aws-neuron
stay managed: /deployment/cohere-platform

Use the recommended generation settings

Hugging Face’s model card recommends temperature=1.0 and top_p=0.95 for generation.

Graduate into workflow-specific docs

Once you have a first response working, move into the agent, architecture, FAQ, and troubleshooting pages as needed.

North Mini Code’s 3B active parameter footprint is why the local paths in this doc set are practical, even though the model has 30B total parameters.