The Canadian AI firm this week introduced North Small Translate 1.0 under a CC BY-NC 4.0 license, making model weights available for download, testing, and research purposes. However, deploying the system in production environments requires a separate commercial agreement.
Cohere positions itself around AI sovereignty for regulated sectors and frames this launch as advancing sovereign AI as a technological reality. In this context, sovereignty refers to maintaining authority over infrastructure location and data visibility. The commercial licensing structure preserves this commitment while keeping enterprises dependent on Cohere's terms. Organizations retain data and infrastructure control but cannot fork the model, develop commercial products from it, or continue operation if renewal terms shift.
Open weights, except for commercial production
North Small Translate operates as an open-weights mixture-of-experts architecture designed for translation work spanning more than 50 languages and regional variants. The system contains 218 billion total parameters, with 25 billion active during operation and supports a 16,000-token context window.
Not all users have the same access to those weights.
According to Cohere, the model enables researchers, developers, and enterprises to evaluate and deploy machine translation while retaining control over their data and infrastructure. This messaging appeals to organizations pursuing sovereign AI approaches, yet the open-weight availability carries a significant restriction: access rights differ based on intended use.
The model is accessible through Cohere's complimentary tier via the Chat V2 API. Those planning non-commercial applications can obtain FP8 weights from Hugging Face under CC BY-NC 4.0 licensing.
Organizations seeking production deployment face different requirements. They must acquire a commercial license and run North Small Translate through Model Vault, Cohere's managed inference service.
Cohere's not the only one drawing a line around open-weight use
Comparable licensing strategies are emerging across the AI industry as companies attach stricter conditions to open-weight releases.
During the previous month, Chinese AI research organization Z.ai published weights for its GLM-5.3 model on Hugging Face. Like Cohere, Z.ai differentiated licensing based on deployment context—a shift from earlier practices. While GLM-5.2 carried the permissive MIT license, GLM-5.3 introduces additional restrictions for certain commercial actors.
These restrictions target companies generating more than $10 billion in aggregate revenue across 12 consecutive months. Should such organizations wish to host GLM-5.3 or modified versions for commercial purposes, they must obtain clearance through Z.ai's security evaluation process.
Z.ai provided no public explanation for reversing course on GLM-5.3, particularly surprising given that GLM-5.2 operated under MIT terms without commercial restrictions. Similarly, Cohere has declined to publicly articulate its reasoning behind restricting North Small Translate's commercial applications.
Cohere, for its part, has been similarly mum about why it made North Small Translate's open weights noncommercial.
Sovereign deployment, with restrictions
Cohere's approach to North Small Translate—offering open weights while restricting commercial deployment—appears inconsistent with its established track record of marketing sovereign AI solutions to enterprise clients.
In June, the company introduced North Mini Code, its initial coding model, as fulfilling developer demands for sovereignty protections comparable to those long demanded by regulated industries. This open-weight model shipped under Apache 2.0 licensing without equivalent commercial-use barriers.
North Small Translate signals a strategic departure for Cohere, exemplifying broader industry trends toward imposing stricter conditions on increasingly sophisticated open-weight systems.
Source: The New Stack