Supported Models
Architecture
Model Vault Encrypted runs on the same isolated, dedicated, Cohere-managed infrastructure as a Standard Vault, and adds on top:
- Confidential computing: inference runs inside hardware-backed trusted execution environments (a confidential VM such as Intel TDX or AMD SEV-SNP on the CPU, and an NVIDIA GPU in confidential-computing mode), so data stays protected in use, not just in transit and at rest. See the Security Model and Confidential Computing Primer.
For how data is protected in transit, at rest, and in use, and how Zero Data Retention (ZDR) is handled, see Encryption & Key Management.
Supported models
Encrypted Vaults run in the United States, with regional serving in the European Union and Canada coming soon (see Regions). Every model below runs on one NVIDIA H100 80GB GPU in confidential-computing mode.
Cohere models
Open models
An Encrypted Vault runs open models from the following families. The model catalogue shows the specific models available when the Encrypted option is selected.
Open models are called through the OpenAI-compatible Chat Completions API. See Calling an Encrypted Vault over the API.
We are expanding this support over time, including larger models, so availability can differ from a Standard Vault. To confirm availability for a specific model, check the model catalogue when creating a vault with the Encrypted option selected, or contact Cohere.
Supported GPUs
Encrypted Vaults run on confidential-computing-capable NVIDIA H100 GPUs. To confirm the options available for your model, capacity, and region, contact Cohere.
Encrypted Vaults are priced per GPU, per hour. See the pricing page.