Model Vault
Product · API / developer platform by Cohere
Dedicated, fully managed model inference platform with optional confidential computing.
Model Vault is a dedicated, fully managed SaaS inference platform for deploying Cohere's frontier models and leading third-party open models. It provides the convenience of an API with the security of private hosting, allowing enterprises to run workloads in isolation without the operational overhead of managing infrastructure. For highly regulated industries, Model Vault Encrypted adds confidential computing, ensuring that AI inference runs fully encrypted inside hardware-secured enclaves. The platform supports regional serving in the United States, European Union, and Canada to meet data sov
- GPU Types
- L4, A100, MI300X, H100, H200, B200, GB200
- Available Regions
- United States, European Union, Canada
- Data Policy
- Zero Data Retention (ZDR) by default
- Third-party models
- DeepSeek, Gemma, GPT-OSS, Qwen, Phi, Nemotron
Overview
Model Vault is a dedicated, fully managed SaaS inference platform for deploying Cohere's frontier models and leading third-party open models. It provides the convenience of an API with the security of private hosting, allowing enterprises to run workloads in isolation without the operational overhead of managing infrastructure. For highly regulated industries, Model Vault Encrypted adds confidential computing, ensuring that AI inference runs fully encrypted inside hardware-secured enclaves. The platform supports regional serving in the United States, European Union, and Canada to meet data sovereignty requirements. It is designed to handle high-throughput agentic workloads with guaranteed performance and zero data retention.
Managed Infrastructure
- Dedicated Inference: Provides a dedicated, fully managed SaaS inference platform for Cohere and leading open models.
- No Shared Resources: Eliminates performance degradation from resource sharing with no multi-tenancy overhead.
- Guaranteed Performance: Runs production workloads without rate limits through dedicated infrastructure.
- Regional Serving: Allows customers to choose the specific region for model serving, including the US, EU, and Canada.
Model Vault Encrypted
- Confidential Computing: Uses hardware-secured enclaves to run fully encrypted AI inference.
- Zero Data Retention: Ensures that prompts and responses are not stored after processing.
- Remote Attestation: Allows cryptographic verification of protection at any time through remote attestation.
- End-to-End Encryption: Protects data from being seen by Cohere or the cloud provider during inference.
Pricing
- L4 24GB GPU (Standard)
- $3.00 per hour
- A100 80GB GPU (Standard)
- $3.50 per hour
- H100 80GB GPU (Standard)
- $6.25 per hour
- H100 80GB GPU (Encrypted)
- $9.00 per hour
- B200 192GB GPU (Encrypted)
- $12.00 per hour
Specifications
- GPU Types
- L4, A100, MI300X, H100, H200, B200, GB200
- Available Regions
- United States, European Union, Canada
- Data Policy
- Zero Data Retention (ZDR) by default
- Third-party models
- DeepSeek, Gemma, GPT-OSS, Qwen, Phi, Nemotron
Added by AntSpace from public sources (cohere.com).