Introducing North Small Translate: A leading sovereign open-weight machine translation model

Today, we're releasing North Small Translate, a mixture-of-experts machine translation model with strong performance across 50+ languages. Across WMT26 benchmarks,¹ North Small Translate achieves an 83.6 score across all languages, outperforming proprietary models like DeepL and Google Translate, as well as open-weight alternatives such as Gemma 4 31B (off), GLM 5.2, and Mistral Large 3.

North Small Translate marks a significant milestone as Cohere's first translation model in the North model family. It builds on our multilingual and translation lineage — from the Tiny Aya model family to Command A Translate — and represents a clear next step in Cohere’s commitment to offering high-quality machine translation wherever it is needed.

Now available for research and non-commercial use under a CC BY-NC 4.0 license, North Small Translate advances Cohere’s mission to make sovereign AI a technological reality.

Visit Hugging Face to download the weights - available in several near-lossless quantizations - explore our HuggingFace space to demo the model, and read our implementation guides .

North Small Translate outperforms similarly sized open-weight models under 1T parameters and API-based translation models in various dimensions of machine translation on average. In WMT model evaluations, North Small Translate leads with an WMT26 All Languages benchmark score of 83.60, compared with 81.56 for Qwen 3.5 397B A17B, 76.50 for GLM 5.2 FP8, 81.37 for DeepL NextGen, 79.46 for Gemma 4 31B (on), and 68.20 for Google Translate. North Small Translate (Agentic) — which can find errors and fix errors in translation — scores even higher, at 84.36.²

North Small Translate was developed in partnership with RWS , an AI solutions company pioneering in language technology and services. Close collaboration with RWS, specifically Language Weaver’s research and science teams along with its language experts, helped shape the model's real-world translation performance throughout development. RWS works with more than 80% of the world’s top 100 brands, empowering the world's most ambitious brands to communicate seamlessly across borders and cultures.

Image 4: Overall WMT score and cost per task breakdown across machine translation models, using pricing per token or character from model providers' websites.

North Small Translate is available today on Hugging Face for non-commercial and research use. Visit our documentation for detailed model specs, deployment guides, and implementation examples to get started. For information about how the model was developed, read our technical report .

Image 1: Overall WMT26 scores across all languages across major models, benchmarked for WMT 2026 using GPT-5.6-Sol as a judge.

North Small Translate performs strongly across 32 high-resource languages and 18 additional languages. North Small Translate is the most consistent performer across the full spread of regions, without the sharp regional drop-offs seen in other models of its size. On average across all languages, it is the best-performing dedicated machine translation model in this evaluation — open or closed.

Image 2: Overall WMT scores by regional average across DeepL, Google Translate, and Gemma 31B (on), and North Small Translate using GPT-5.6-Sol as a judge.