Cohere releases North Small Translate machine translation model

North Small Translate machine translation open-weight model mixture-of-experts AI sovereignty

Cohere has released North Small Translate, a mixture-of-experts (MoE) model designed specifically for machine translation. The model features 218 billion total parameters, with 25 billion active parameters, and supports over 50 languages. It includes a 16,000-token context window for both input and output.

In benchmark testing using the WMT26 metric, North Small Translate achieved an average score of 83.60. When utilizing a multi-pass agentic variant to correct errors, the score increased to 84.36. These results outperformed other models in the evaluation, including DeepL NextGen, Google Translate, and various open-source models.

While Cohere provides the model weights under a CC BY-NC 4.0 license for research and non-commercial use via Hugging Face, commercial deployment is restricted. Enterprises intending to use the weights in production must purchase a commercial license and deploy the model through Cohere’s managed inference platform, Model Vault.

Entities

Cohere

Sources

Cohere presenta North Small Translate, modello MoE open-weight da 218 miliardi di parametri per la traduzione automatica – iHAL [ihal.it]

7 days ago

Cohere says AI sovereignty matters — except it just gated its new translation model for commercial use [thenewstack.io]

6 days ago