What is North Small Translate?
North Small Translate is a machine translation model from Cohere's North model family. Its purpose is straightforward: translate business and operational content between supported languages while giving organizations options for hosted API access, private infrastructure, or self-hosting with open weights.
Cohere documents the model as a mixture-of-experts, or MoE, system with 218 billion total parameters and 25 billion active parameters. In an MoE model, only a portion of the full parameter set is activated for a particular input. That distinction helps explain the model's positioning: the total model is very large, but each request does not use all 218 billion parameters at once.
The model is designed for enterprise translation workloads such as internal documentation, technical manuals, safety procedures, company communications, customer-support content, and localization. It is not primarily positioned as a general-purpose chatbot, software-development model, image model, or audio model.
Where it fits in Cohere's lineup
North Small Translate sits within Cohere's North family, alongside a broader ecosystem that includes generation, retrieval, reranking, document-processing, and multilingual model capabilities. Unlike a general assistant, its defining specialization is translation.
Cohere lists reasoning, tool use, citations, safety modes, and structured outputs among the documented capabilities for the model. These features can be useful in an enterprise workflow, for example when translation is combined with document processing or an application needs structured results. However, the model's central value remains multilingual text transformation rather than open-ended reasoning or agentic task completion.
That positioning matters when evaluating it against other model types. A general-purpose language model may be a better choice for complex writing, software generation, or extended conversational work. North Small Translate is more directly suited to organizations that need a dedicated translation component and want control over where that component runs.
Languages and translation capabilities
Cohere states that North Small Translate supports English and more than 50 languages and locale variants. The documented tier-one language group includes Modern Standard Arabic, German, French, Japanese, Korean, Russian, and Ukrainian. The broader language list also includes Spanish, Chinese, Hindi, Portuguese, Turkish, Vietnamese, Thai, Urdu, and Indonesian.
For an enterprise, broad language and locale coverage can reduce the need to maintain separate translation systems for different regions. Potential uses include translating product and policy documentation, preparing localized support material, converting safety instructions for international teams, and assisting multilingual knowledge-management workflows.
The supplied documentation does not provide independent benchmark results or detailed quality scores for each language pair. Translation quality can vary by language, domain, terminology, and document style, so organizations should validate the model using their own representative content before adopting it for high-impact or regulated workflows.
Context window and output limits
North Small Translate has a documented 16K-token context window and a maximum output length of 16K tokens. Tokens are the text units processed by the model; they do not correspond exactly to words, so the practical amount of text depends on the languages and formatting involved.
The 16K context makes the model suitable for substantial documents or larger sections of manuals and internal material, although users still need to account for both the source text and the requested translation within the available request capacity. A long document may need to be divided into sections, particularly when the application also includes instructions, terminology, metadata, or other context.
The maximum output is also documented as 16K tokens. This is a capacity limit, not a guarantee that a translation will use the entire allowance. Output length normally depends on the source language, target language, formatting, and the amount of explanatory or structured content requested by the application.
Deployment options and open weights
North Small Translate is available through Cohere's Chat V2 API, with hosted access available on the free tier until applicable rate limits are reached. This provides a way to evaluate or integrate the model without immediately operating the model on company infrastructure. The free access should not be confused with unlimited production usage: rate limits apply, and Cohere's general trial terms distinguish evaluation access from production or commercial use.
Cohere also provides open weights for non-commercial use under the Creative Commons Attribution-NonCommercial 4.0 license. The documented formats are W4A16, FP8, and BF16. These formats represent different efficiency and numerical-precision trade-offs, and they affect the hardware needed for deployment.
Cohere suggests the following hardware configurations:
- W4A16: two H100 GPUs or one B200 GPU.
- FP8: four H100 GPUs or two B200 GPUs.
- BF16: eight H100 GPUs or four B200 GPUs.
These requirements show why the model is aimed at organizations with substantial infrastructure or a need for managed private deployment. The active-parameter count is smaller than the total parameter count, but self-hosting still requires significant GPU capacity.
Pricing and commercial access
For Cohere's hosted API, the supplied documentation describes North Small Translate as available for free until applicable rate limits are reached. No separate token-based input or output price is published in the supplied research.
Cohere's Standard Model Vault pricing lists the North Small Translate XL deployment at 17.50 US dollars per instance-hour. This is an infrastructure-style hourly price, not a per-token price and not a monthly subscription. Access may also depend on Cohere's enterprise availability and deployment process.
The pricing models serve different needs. API access is simpler for teams that want to test or integrate translation without managing GPUs. Model Vault or another private deployment is more relevant when data residency, isolation, governance, or infrastructure control is more important than the lowest possible setup cost. The open-weight license must also be considered: the supplied research describes it as non-commercial, so commercial self-hosting requires checking Cohere's applicable terms and obtaining the appropriate arrangement.
Modalities, tools, and structured output
North Small Translate uses text input and produces text output. It does not generate images, audio, or video, and the supplied specifications do not identify image, audio, or video input.
The model documentation lists tool use, citations, safety modes, and structured outputs as supported capabilities. Tool use can allow an application to connect the model to external functions or workflow steps, although the model itself should not be treated as a web-search system: the supplied specifications mark web-search support as unavailable. Structured-output support can help applications request predictable response formats, but the research does not verify a separate JSON mode, so those two concepts should not be treated as identical.
Reasoning is listed among the provider-documented capabilities, but North Small Translate remains a translation-focused model. Its reasoning support is most relevant when an application needs translation-related instructions, terminology handling, citations, or workflow decisions. It should not automatically be assumed to match a specialized reasoning model on difficult mathematical, analytical, or coding tasks.
Strengths and limitations
Main strengths
- Specialized translation focus: The model is built for multilingual translation rather than requiring a general assistant to handle translation as one task among many.
- Broad language coverage: It supports English and more than 50 languages and locale variants, including major European, Asian, and Middle Eastern languages.
- Large documented capacity: A 16K-token context window and 16K-token maximum output support substantial translation requests.
- Deployment flexibility: Organizations can use Cohere's API, pursue private deployment, or use open weights for non-commercial self-hosting.
- Enterprise workflow features: Provider-documented tool use, citations, safety modes, and structured outputs can support translation inside larger document and business processes.
Main limitations
- Hardware demands: Self-hosting requires multiple high-end GPUs under the configurations documented by Cohere.
- License restrictions: The open-weight release is described as CC BY-NC 4.0, which limits straightforward commercial use.
- Limited public pricing detail: Hosted token prices are not published in the supplied research, while Model Vault pricing is expressed per instance-hour and may require enterprise access.
- Not a general-purpose assistant: Translation is the primary use case; users seeking image generation, audio, video, broad coding, or consumer chat should consider another type of model.
- Quality still requires validation: No supplied benchmark establishes uniform performance across all supported languages, domains, or terminology-heavy documents.
When to choose North Small Translate
Choose North Small Translate when translation is a central requirement and your organization needs multilingual coverage, enterprise integration, or control over deployment. It is especially relevant for internal documentation, technical and safety material, customer-support localization, international operations, and workflows where sensitive documents should remain within private or controlled infrastructure.
The hosted API is the practical starting point when a team wants to test the model or avoid GPU management. A private or Model Vault deployment is more appropriate when data residency, isolation, governance, or predictable infrastructure control outweighs the simplicity of a hosted API. Non-commercial research and evaluation teams may also find the open weights useful, provided the hardware and license terms fit their project.
Another model type may be more appropriate when translation is only a small part of a broader task. A general-purpose language model is likely a better fit for open-ended conversation, complex coding, extensive document drafting, or broad agent workflows. A dedicated speech model is preferable for audio transcription, while a vision or generative media model is required for image and video tasks. Even within Cohere's catalog, the Command, Embed, Rerank, Parse, and Aya families address different generation, retrieval, document, and multilingual needs; North Small Translate should be selected specifically when translation is the main workload.
Overall assessment
North Small Translate is a specialized enterprise translation model with an unusually large total parameter count, broad language coverage, and multiple deployment paths. Its strongest practical distinction is not that it replaces every kind of language model, but that it gives organizations a translation-focused option that can be accessed through Cohere's API or deployed in more controlled environments.
The main trade-off is operational: the model's open-weight and private-deployment flexibility comes with substantial hardware requirements, while commercial access and pricing may require an enterprise process. For teams that need multilingual translation at scale and can meet those operational requirements, it is a focused alternative to using a general-purpose model for every translation task.