Cohere Releases North Small Translate 1.0 with Restrictive Open-Weight License
The 218-billion parameter translation model is open for research, but commercial production requires a paid agreement.
Cohere has launched North Small Translate 1.0, a high-capacity machine translation model released with open weights for research and evaluation. While the model is accessible to the public, its licensing terms create a strict divide between academic study and commercial application.
The model utilizes a mixture-of-experts (MoE) architecture, featuring 218 billion total parameters with 25 billion active parameters. According to technical specifications, North Small Translate 1.0 supports machine translation across 50 languages and provides a 16,000-token context window for both input and output. Non-commercial weights are currently hosted on Hugging Face under the CC BY-NC 4.0 license, which explicitly prohibits commercial use.
A Shift in Licensing Strategy
This release is framed by Cohere as part of a broader mission to make "sovereign AI a technological reality," focusing on providing regulated industries with greater control over their data and infrastructure. However, the choice of the CC BY-NC 4.0 license marks a departure from some of the company's previous openness. In June 2026, Cohere released North Mini Code under the permissive Apache 2.0 license, which allowed for much broader commercial adoption without the restrictions seen in the new translation model.
The Commercial Gate
By gating the model's production use, Cohere is following a growing industry trend where AI labs provide weights for transparency and research while reserving monetization for enterprise deployments. For organizations wishing to move beyond evaluation and into production, a commercial agreement is required. Cohere offers Model Vault, a fully managed SaaS inference platform designed for deploying models in isolated, single-tenant setups, as a path for secure enterprise implementation.
Industry Implications
This strategy reflects a balancing act between community engagement and revenue protection. By allowing researchers to download and study the weights, Cohere benefits from community feedback and academic validation. Simultaneously, by restricting commercial use, the company ensures that high-capability models remain a paid product in the competitive enterprise market. This approach mirrors moves by other labs, such as Z.ai with GLM-5.3, which have similarly attached restrictive conditions to their open-weight releases.
What to Watch
As more "sovereign AI" tools enter the market, the tension between open-weight availability and commercial restrictions will likely intensify. Observers will be watching to see if other Cohere releases return to permissive licenses like Apache 2.0 or if the restrictive CC BY-NC 4.0 model becomes the new standard for the company's high-parameter releases. It remains to be seen how the research community will react to the limitations on production deployment for a model of this scale.