Cohere Launches North Small Translate AI Model
Cohere has released North Small Translate, an open-weight model that outperforms major proprietary translation systems while significantly lowering operational costs for enterprises.
Cohere, in partnership with AI solutions firm RWS, has released North Small Translate, a mixture-of-experts machine translation model supporting over 50 languages. Marking the first translation model in Cohere's North family, the open-weight system is available on Hugging Face under a CC BY-NC 4.0 license for research and non-commercial use, while commercial access is offered via RWS's Language Weaver platform.
The model achieves a score of 83.60 on the WMT26 All Languages benchmark, outperforming Qwen 3.5 397B A17B at 81.56, DeepL NextGen at 81.37, Gemma 4 31B (on) at 79.46, GLM 5.2 FP8 at 76.50, and Google Translate at 68.20. An error-correcting agentic version of the model scores even higher at 84.36. Regionally, the standard model beats Gemma 4 31B (on) in Europe with a score of 82.17 versus 72.73, while running nearly even in South Asia at 86.16 versus 88.04. It also beats DeepL NextGen across all tested non-European regions, leading by up to 10 points in South Asia and MENA.
In terms of speed, North Small Translate delivers up to 1.4 times the throughput of Gemma 4 31B TP1, generating 112 tokens per second at low concurrency compared to Gemma's 81. For long-context tasks, such as translating two book chapters in a single call, Cohere's model scored 48.9, easily beating Google Translate's 21.3 and Gemma 4's 19.4.
For enterprise deployments, the model offers significant cost advantages. It achieves an 80.1 score at an average cost of $0.000676 per task using 661 tokens. This is substantially cheaper than Gemini 3.1 Pro Preview, which costs $0.038928 per task, as well as Qwen 3.5 397B A17B at $0.004525 and Cohere's own Command A+ at $0.005158.
This is our own summary of reporting by Cohere Blog
