Cohere Model Vault (Encrypted)

Declared: Attested E2EE

Dedicated single-tenant confidential deployments priced per GPU-hour, reached through a client-side OHTTP proxy

Cohere Model Vault (Encrypted) (Cohere, Toronto, Canada) is a dedicated single-tenant confidential deployment, not a shared per-token API. Per the provider's supported-models doc, every listed model runs on one NVIDIA H100 80GB GPU in confidential-computing mode, and encrypted vaults are priced per GPU, per hour. All traffic must go through the client-side Cohere OHTTP proxy (ghcr.io/cohere-ai/tng-ingress or the conseel Python wrapper), which verifies composite CPU+GPU attestation and encrypts end to end with OHTTP: Intel TDX tokens via Intel Trust Authority (a 'passport') or AMD SEV-SNP appraised locally by the proxy, plus NVIDIA NRAS for the GPU. OpenAI-compatible /v1 serves the open model families (Gemma, GPT-OSS, Phi, Qwen, Ministral/Devstral); Cohere families (Embed, Rerank, Parse, Compass, North Mini Code, Transcribe, customized) use the Cohere /v2 API. The US region is available now, with EU and Canada announced. Attestation is exchanged per customer deployment through the proxy, so this directory cannot monitor it: TEE, E2EE and hardware claims here are provider claims.

HQToronto, Canada

Models2

From$4.25/GPU-h (declared, promo — list $9.00/GPU-h until 2027-03-31)

Organization rolesStorefront · Workload operator

Trust chainIntel (TDX) + AMD (SEV-SNP) + NVIDIA (GPU CC) · CPU TEE in path: yes · physical attacks

E2EE (declared)OHTTP

Advertised ChannelAttested E2EE

Post-Quantum KEXNot measured

API Features

OpenAI Compatible✓

Streaming—

Function Calling—

Batch Inference—

Vision—

Embeddings✓

Models & Pricing

Rates are exactly what the provider source published, kept in the recorded currency; USD eligibility is evaluated from fresh observations. Declared conditions and undisclosed terms are listed, not verified.

Model / source ID	Family	Context	Max Out	Offerings	Attestation	Observed route
Model Vault Encrypted H100 80GBmodel-vault-encrypted-h100-80gb	Model Vault	—	—	per hour`model-vault-encrypted-h100-80gb` $4.25 per compute hour declared declared price source · observed 2026-11-09 · price source Promotional rate — list $9.00 through 2027-03-31 Dedicated single-tenant vault, one H100 80GB GPU in confidential-computing mode. Promotional $4.25 per GPU-hour through 2027-03-31 — the effective price while the promo runs; $9.00 per GPU-hour on-demand list price. Reserved capacity: monthly and annual commit pricing available.	No current check	Not identified
Model Vault Encrypted B200 192GBmodel-vault-encrypted-b200-192gb	Model Vault	—	—	per hour`model-vault-encrypted-b200-192gb` $8.00 per compute hour declared declared price source · observed 2026-11-09 · price source Promotional rate — list $12.00 through 2027-03-31 Dedicated single-tenant vault; the B200 192GB configuration is listed on the pricing page. Promotional $8.00 per GPU-hour through 2027-03-31 — the effective price while the promo runs; $12.00 per GPU-hour on-demand list price. The supported-models doc names H100 80GB for encrypted vaults; B200 192GB appears on the pricing page. Reserved capacity: monthly and annual commit pricing available.	No current check	Not identified