Sakhanda Wire
NVDA $154.00 -1.37% MSFT $506.28 -2.26% GOOGL $280.87 -1.00% META $620.34 -0.97% AMZN $256.88 -0.34%
← Back to the news

Cohere releases 218B-parameter translation model for 50 languages

Cohere has released North Small Translate, an open-weight machine-translation model developed with Cohere Labs and RWS. The sparse Mixture-of-Experts system contains 218B total parameters, with 25B active for each token, and supports 50 languages from Albanian to Vietnamese. On Cohere’s WMT26 evaluation, it achieved an average score of 83.60 across all languages; an Agentic version using multiple passes reached 84.36.

Cohere’s comparisons put the standard model ahead of DeepL NextGen, Google Translate, Qwen 3.5 397B A17B and other systems. The company cautions that these results were produced in-house and judged by GPT-5.6-Sol, so independent WMT26 results have not yet confirmed them. In tests cited by Cohere, North Small Translate also delivered higher throughput than Gemma 4 31B and performed better on translations of two book chapters.

The model accepts 16K input and output tokens. It can be accessed through Cohere’s API free of charge until rate limits apply, self-hosted for non-commercial use, or licensed commercially. Cohere publishes checkpoints for Blackwell, Hopper and NVFP4 configurations; the 4-bit version runs on one B200 or two H100 GPUs.

Written by Sakhanda Wire AI from reporting by MarkTechPost. Facts, quotes and full context are in the original article below.

Originally published by MarkTechPost on

Read the original on MarkTechPost ↗

The reporting and the image belong to MarkTechPost and are credited and linked above; the text on this page is Sakhanda Wire's own summary.

← Back to the news

More stories

All the latest news