Cohere Adds Open Models to Model Vault

Cohere has launched third-party open models in Model Vault and Model Vault Encrypted, letting enterprises run Cohere models and models from other providers on a managed inference platform. Prompts and outputs are not retained by default in Model Vault. The service is available in the United States, European Union and Canada; Model Vault Encrypted is available in the United States, with EU and Canada support coming soon.

Model Vault runs workloads on single-tenant infrastructure dedicated to one organization. Customers choose a supported region, and Cohere manages the infrastructure and model serving. Model Vault Encrypted adds confidential computing (processing data while it remains encrypted): prompts and outputs stay encrypted inside isolated environments while a model processes them. Cohere says neither it nor cloud providers can see that data.

The launch includes open models such as DeepSeek, GLM, MiMo, Gemma, Nemotron, GPT-OSS, Phi and Hy3, alongside Cohere's Command, Embed, Rerank, Parse, Transcribe and Translate families. The availability table lists 15 model families: 14 are available in Model Vault, while Command is not listed for the Encrypted tier. Encrypted access is listed for Embed, Rerank, Parse, Transcribe, Translate, Gemma, GPT-OSS and Phi. Cohere also names DeepSeek V4.1 Flash, GLM 5.3 Flash and MiMo V2.6 Pro among the open models it offers; it describes DeepSeek V4.1 Flash as the most-used open model on OpenRouter.

Cohere describes sovereign inference as customer control over where data is stored, who can access it, which models are used and how AI workloads run. It says Model Vault uses an isolated tenant and in-region processing, while the Encrypted tier is intended for sensitive work where nobody outside the organization should see prompts or outputs. RWS's Language Weaver says it uses Model Vault to run models, including its LW Pro model, while controlling where customer data goes; RWS executive Dragos Munteanu said Cohere handles hosting and scaling.

Model Vault pricing is based on GPU type, with hourly on-demand pricing and promotions; reserved capacity and monthly and annual commitments are available through sales. Cohere says it plans to add regions, extend Encrypted regional service to Canada and the EU, support larger models in Encrypted, and add models across regions over the coming months.

Cohere has launched third-party open models in Model Vault and Model Vault Encrypted, letting enterprises run Cohere models and models from other providers on a managed inference platform. Prompts and outputs are not retained by default in Model Vault. The service is available in the United States, European Union and Canada; Model Vault Encrypted is available in the United States, with EU and Canada support coming soon.

Model Vault runs workloads on single-tenant infrastructure dedicated to one organization. Customers choose a supported region, and Cohere manages the infrastructure and model serving. Model Vault Encrypted adds confidential computing (processing data while it remains encrypted): prompts and outputs stay encrypted inside isolated environments while a model processes them. Cohere says neither it nor cloud providers can see that data.

The launch includes open models such as DeepSeek, GLM, MiMo, Gemma, Nemotron, GPT-OSS, Phi and Hy3, alongside Cohere's Command, Embed, Rerank, Parse, Transcribe and Translate families. The availability table lists 15 model families: 14 are available in Model Vault, while Command is not listed for the Encrypted tier. Encrypted access is listed for Embed, Rerank, Parse, Transcribe, Translate, Gemma, GPT-OSS and Phi. Cohere also names DeepSeek V4.1 Flash, GLM 5.3 Flash and MiMo V2.6 Pro among the open models it offers; it describes DeepSeek V4.1 Flash as the most-used open model on OpenRouter.

Cohere describes sovereign inference as customer control over where data is stored, who can access it, which models are used and how AI workloads run. It says Model Vault uses an isolated tenant and in-region processing, while the Encrypted tier is intended for sensitive work where nobody outside the organization should see prompts or outputs. RWS's Language Weaver says it uses Model Vault to run models, including its LW Pro model, while controlling where customer data goes; RWS executive Dragos Munteanu said Cohere handles hosting and scaling.

Model Vault pricing is based on GPU type, with hourly on-demand pricing and promotions; reserved capacity and monthly and annual commitments are available through sales. Cohere says it plans to add regions, extend Encrypted regional service to Canada and the EU, support larger models in Encrypted, and add models across regions over the coming months.