Last week, enterprise AI vendor Cohere introduced North Small Translate, an open-weight mixture-of-experts machine translation system spanning 50 languages. The model is available for noncommercial development under CC BY-NC 4.0 licensing, while commercial deployment occurs through Cohere's Model Vault, a managed inference service. The company frames the offering as part of its sovereign AI initiative, targeting organizations seeking greater autonomy over model execution locations and data governance practices.

North Small Translate extends Cohere's existing translation portfolio, which encompasses the Tiny Aya and Command A Translate model families. According to Cohere's claims, the system outperforms comparably sized open-weight models under 1T parameters and API-driven translation services across multiple machine translation metrics. The model's efficiency stems from its non-reasoning design—it operates through learned statistical patterns rather than step-by-step logical inference, thereby reducing token consumption, according to Cohere co-founder Nick Frosst.

Translation gaps persist across most global languages

We spent nine years scaling an architecture invented to fix translation, and machine translation is still broken for most of the world's languages. General-purpose models get you most of the way and then stop. The next phase of enterprise AI in this space is smaller, more specialized, and runs inside your own walls.

Nick Frosst, Cohere co-founder

On WMT26 benchmarks, Cohere reports North Small Translate achieving an All Languages score of 83.60, surpassing Qwen 3.5 397B A17B at 81.56, GLM 5.2 FP8 at 76.50, DeepL NextGen at 81.37, Gemma 4 31B at 79.46, and Google Translate at 68.20. The model incorporates 218 billion total parameters with 25 billion active, delivering a smaller computational and memory footprint relative to competing systems. Parameter disclosure inconsistencies across vendors complicate direct model-to-model comparisons in this category.

Long-document translation remains problematic

Machine translation allows documents to be translated from one language to another automatically. With current solutions, long documents start to fall apart. Google Translate scores 21.3 on our long-context test, Gemma 4 31B 19.4; we score 48.9. That's [for example] a safety manual that reads fine on page one… and has drifted by page ten. The other risk is where the text goes. Once you push HR policies or regulated documents through a third-party API, that data has left your building, and necessarily that means your control over it is diminished.

Nick Frosst

Frosst attributes North Small Translate's superior performance on complex enterprise translation assignments to its capacity for handling substantial volumes of confidential materials. The system covers 50 languages—32 designated as high-resource and 18 others—alongside translation-workflow capabilities including structured translations for Markdown and JSON formats, instruction-following for tone and format preferences, and terminology guides for vocabulary standardization.

Multi-pass workflows enhance translation accuracy

North Small Translate works with a multi-pass workflow. The model translates, reviews its own output, finds errors, and fixes them – and this is the same loop we used in training. We ship both because standard is one pass and built for volume, while the agentic [version] spends more tokens for 84.36 against 83.60 on WMT26. That difference ends up being worth it when the document is a contract or a safety procedure, for instance, but in other cases you'd rather optimize for efficiency.

Nick Frosst

The model leverages the same architectural foundation as earlier Cohere systems but achieves performance gains through post-training refinements, including reinforcement learning and translation-specific datasets. Across the translation market, Frosst observes that generative machine translation models deliver superior quality and steerability—encompassing tone and formatting control—though they typically command higher costs than Neural Machine Translation systems prevalent in commercial deployments.

Model 'steerability' drives suggesting language tone and formatting

RWS, a language technology and AI solutions provider, partnered with Cohere during development. Contributions from RWS's Language Weaver research teams and language specialists informed the model's real-world translation capabilities throughout the project.

Developers can obtain model weights at no cost for noncommercial applications in three quantization formats. A Hugging Face Space and API option serve users without adequate hardware resources.

Source: The New Stack