Cohere has released its new translation model, North Small Translate 1.0, with a twist: the weights are open for research and non-commercial use, but commercial deployment requires a paid license. The model supports over 50 languages and outperforms competitors like DeepL and Google Translate, yet its licensing terms create a divide between researchers and enterprises.
Open weights, but not for production
The model, a mixture-of-experts architecture with 218 billion parameters (25 billion active), is available under a CC BY-NC 4.0 license—meaning it can be downloaded, studied, or used for non-commercial purposes. Researchers and developers can access the FP8 weights on Hugging Face, but enterprises wanting to deploy it in production must negotiate a commercial license through Cohere’s Model Vault, a fully managed inference platform.
Cohere frames this as a way to balance openness with control. The company’s free tier via its Chat V2 API lets users test the model, while the commercial route ensures enterprises retain sovereignty over their data and infrastructure. But the restriction on commercial use without a license is a sharp contrast to its earlier approach with North Mini Code, which launched under the permissive Apache 2.0 license with no such barriers.
This split reflects a growing trend in AI licensing. Last month, Z.ai, a Chinese AI lab, released weights for its GLM-5.3 model under similar conditions—open for research but requiring a security review and commercial restrictions for large companies. Unlike its predecessor, GLM-5.2, which used the permissive MIT license, GLM-5.3 now imposes new rules for commercial deployment, particularly for firms with over $10 billion in annual revenue.
The reasoning behind these shifts remains unclear. Neither Cohere nor Z.ai has explained why they’re tightening commercial access, despite earlier models offering broader permissiveness. The move suggests a broader industry shift: as open-weight models grow more capable, companies are redefining what “open” means.
A model built for sovereignty—but with limits
Cohere has positioned itself as a leader in sovereign AI, catering to regulated industries that prioritize data control and on-premises deployment. North Small Translate aligns with that vision, allowing enterprises to evaluate the model locally while keeping their infrastructure and data secure. But the commercial license requirement complicates that promise.
Read Also: GPT Images 2.5 offers precise, isolated image edits
Enterprises can’t fork the model, build derivative products, or lock in usage if terms change at renewal. The open weights serve as a demo, flexible for testing, but not a foundation for independent operation. This approach keeps Cohere’s influence intact, even as it markets the model as a tool for sovereign control.
For researchers and smaller teams, the CC BY-NC 4.0 license is generous. They can experiment, study, or deploy the model for non-commercial projects without restrictions. But for companies scaling translation systems, the cost of a commercial license becomes a hurdle.
This isn’t just about revenue. It’s about maintaining influence. By controlling commercial access, Cohere ensures that even as the model spreads, its own ecosystem, Model Vault, APIs, and support, remains central. The open weights are a carrot, but the commercial terms are the stick.
Why the shift now?
Cohere’s decision to restrict commercial use stands out when compared to its earlier releases. North Mini Code, for example, launched under Apache 2.0 with no such limitations. That model was designed for developers who needed flexibility, and the permissive license reflected that. North Small Translate, however, targets enterprises with stricter compliance needs.
The difference may come down to risk. Translation models handle sensitive data, contracts, legal documents, medical records, and enterprises in regulated industries demand ironclad guarantees. An open license could lead to unauthorized forks or misuse, undermining Cohere’s sovereignty pitch. By requiring a commercial agreement, the company ensures it can monitor usage, enforce terms, and maintain control over how the model is deployed.
