{"version":1,"type":"story","url":"https://digestai.news/story/cohere-releases-new-translation-model-with-218b-parameters","json":"https://digestai.news/story/cohere-releases-new-translation-model-with-218b-parameters.json","markdown":"https://digestai.news/story/cohere-releases-new-translation-model-with-218b-parameters.md","slug":"cohere-releases-new-translation-model-with-218b-parameters","headline":"Cohere Releases New Translation Model with 218B Parameters","summary":"Cohere has unveiled North Small Translate, a new machine translation model that boasts 218 billion total parameters and 25 billion active ones. This model is capable of translating across 50 languages, from Albanian to Vietnamese. On the WMT26 evaluation, it achieved an impressive score of 83.6, outperforming DeepL and Google Translate. The model can be deployed via Cohere's API or self-hosted for non-commercial use. It follows previous models in Cohere’s North family such as Tiny Aya and Command A Translate, which were developed with input from RWS scientists. The architecture of the model is a decoder-only sparse Mixture-of-Experts (MoE) Transformer, with 128 experts activated per token. Cohere emphasizes that this model addresses issues of global communication sovereignty in organizations. The model's throughput and cost efficiency are also highlighted, making it competitive with other translation tools like Google Translate and Gemini 3.1 Pro Preview. Overall, North Small Translate is a significant advancement in the field of machine translation models from Cohere.","keyPoints":["Cohere releases North Small Translate","Model scores 83.6 on WMT26 across all languages","Active parameters: 25B out of 218B"],"whyItMatters":"North Small Translate represents a significant step forward in machine translation, particularly for organizations needing accurate global communication solutions.","category":{"slug":"models","name":"Generative AI & Models","url":"https://digestai.news/category/models"},"entities":{"companies":["Cohere","RWS"],"models":["North Small Translate"],"people":[]},"firstPublishedAt":"2026-09-11T06:57:22Z","updatedAt":"2026-09-11T06:57:22Z","sourceCount":2,"hasPrimarySource":false,"sources":[{"outlet":"MarkTechPost","title":"Cohere Releases North Small Translate: A 218B MoE Translation Model That Scores 83.6 on WMT26 Across 50 Languages","url":"https://marktechpost.com/2026/09/10/cohere-releases-north-small-translate-a-218b-moe-translation-model-that-scores-83-6-on-wmt26-across-50-languages","publishedAt":"2026-09-11T06:57:22Z","type":"press","primary":false,"lead":true},{"outlet":"Unite.AI","title":"Cohere Debuts Open-Weight 218B Mixture-of-Experts Machine Translation Model","url":"https://unite.ai/cohere-debuts-open-weight-218b-mixture-of-experts-machine-translation-model","publishedAt":"2026-09-10T15:27:02Z","type":"press","primary":false,"lead":false}],"sourceNotes":null,"discussions":[],"thread":null,"cite":{"text":"Digest AI, \"Cohere Releases New Translation Model with 218B Parameters\", 11 September 2026, https://digestai.news/story/cohere-releases-new-translation-model-with-218b-parameters","publisher":"Digest AI","title":"Cohere Releases New Translation Model with 218B Parameters","datePublished":"2026-09-11T06:57:22Z","url":"https://digestai.news/story/cohere-releases-new-translation-model-with-218b-parameters"},"generatedBy":"Written by Digest AI's editorial model from the linked sources; the sources are the record.","license":"Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse"}