North Small Translate

Try in Playground

North Small Translate is a mixture-of-experts (MoE) model purpose-built for machine translation. It has 218 billion total parameters, with 25 billion active parameters, and a 16K context length.

The model provides flexible deployment options for research and enterprise use. Its weights are openly available in W4A16, FP8, and BF16 under the Creative Commons Attribution-NonCommercial 4.0 license for non-commercial use.

North Small Translate is well-suited for:

• Knowledge management: Translate internal documentation, wikis, standard operating procedures, technical manuals, historical reports, and institutional knowledge.
• Safety and operations: Translate safety manuals, emergency procedures, maintenance instructions, manufacturing procedures, incident reports, field-service documentation, and operational alerts.
• Internal communication: Translate company announcements, leadership communications, HR policies, employee handbooks, benefits information, onboarding materials, newsletters, and intranet content.
• Localization and customer support: Translate product content and customer communications while retaining control over where the model and data are hosted.

North Small Translate supports English and more than 50 language and locale variants. Its tier-one languages are Modern Standard Arabic, German, French, Japanese, Korean, Russian, and Ukrainian.

North Small Translate is available on the free tier through the Chat V2 API. Ask the model to translate the supplied content into a target language:

from cohere import ClientV2

co = ClientV2(api_key="<YOUR_API_KEY>")

target_language = "French"
content = "Enterprises need accurate translations of business-critical documents."

response = co.chat(
    model="north-small-translate-1-0",
    messages=[
        {
            "role": "user",
            "content": f"Translate everything that follows into {target_language}:\n\n{content}",
        }
    ],
)

print(response.message.content[0].text)

• Cohere deployment (free tier - Cohere API): Try the model on the free tier through the Chat V2 API. Usage is subject to limitations set out in the applicable commercial agreements with customers, including Cohere's Usage Policy.
• Open weights deployment: Download the NVFP4 W4A16 model from Hugging Face for non-commercial use under the CC BY-NC 4.0 license.
• Hardware: Suggested hardware by quantization format:
  • W4A16: Two H100s or one B200
  • FP8: Four H100s or two B200s
  • BF16: Eight H100s or four B200s

Machine translation uses software to convert text from one language into another. A purpose-built generative model can act as the translation engine in an application, workflow, or translation platform.

A model gives developers control over integration, prompting, deployment, data residency, and customization. Self-hosted and private deployments can also support sovereignty requirements and reduce dependence on a fixed translation platform.

A translation platform wraps a model with features such as collaboration, content management, project management, and formatting. Use a platform when those workflow features are more important than direct control over the model.

• EU AI Act Article 53(1)(d):

  Public Summary of Training Content