Cohere Gates Commercial Use of North Small Translate 1.0 Amid Sovereignty Push
The release of a 218B parameter translation model under a non-commercial license sparks debate over Cohere's commitment to AI autonomy.
Cohere has released North Small Translate 1.0, a high-capacity translation model available for research but restricted for business use. The move creates a visible tension between the company's licensing strategy and its public advocacy for AI sovereignty.
North Small Translate 1.0 utilizes a sparse Mixture-of-Experts (MoE) architecture featuring 218 billion total parameters, with 25 billion active parameters per token. The system supports 50 languages and provides a 16K input and 16K output context window. Cohere reports a WMT26 score of 83.60, a figure that rises to 84.36 when the model is deployed within an agentic multi-pass translation workflow.
The Licensing Gap
While the model weights are accessible, they are released under the CC BY-NC 4.0 license. This specific license explicitly prohibits commercial application, meaning any entity wishing to use the model for profit must enter into a separate agreement via Cohere's "Model Vault."
This follows the release of Command A Translate in August 2025, as Cohere continues to expand its portfolio of specialized translation tools. By using the North MoE architecture, the company aims to balance high-end performance with computational efficiency, though the access path remains strictly controlled for paying customers.
Sovereignty vs. Gating
This restrictive licensing approach stands in contrast to Cohere's positioning as a champion of AI sovereignty. The company has frequently argued that nations and enterprises should maintain the autonomy to deploy and control their own AI infrastructure to avoid the risks of vendor lock-in. However, by gating the commercial use of open weights, Cohere effectively requires users to return to a vendor-controlled environment for production scaling.
Industry analysts suggest that if sovereignty implies the freedom to deploy AI independently, forcing commercial users into a proprietary vault undermines that philosophy. The move reinforces the very dependency on a single provider that Cohere's branding claims to oppose, creating a paradox where the weights are "open" but the utility is locked.
What's Next
It remains to be seen if Cohere will offer more flexible licensing for the North series or if the Model Vault will become the standard gateway for all its high-parameter MoE releases. As enterprises seek to reduce reliance on closed-API providers, the industry will be watching whether Cohere's definition of sovereignty includes the right to commercially exploit open weights without restrictive vendor contracts. The resolution of this tension will likely signal whether Cohere views sovereignty as a technical capability or a marketing pillar.