Cohere announced on November 3, 2026 that Model Vault and Model Vault Encrypted now run open models from other companies, not only its own models. The change lets organizations choose the most suitable model for each task while keeping the managed, isolated infrastructure offered by the platform.
According to Cohere, Model Vault provides the full inference stack on an isolated instance dedicated to each organization, with zero retention of prompts and outputs by default. Model Vault Encrypted goes further: data stays encrypted during processing, through confidential computing, so that neither Cohere nor the cloud provider would have access to the content sent or received.
Which models are included
In the standard version, Model Vault now runs third-party models such as DeepSeek, GLM (Z.ai), MiMo (Xiaomi), Gemma (Google), Nemotron (NVIDIA), GPT-OSS (OpenAI), Phi (Microsoft) and Hy3 (Tencent), alongside Cohere's own models, including Command, Embed, Rerank, Parse, Transcribe and Translate. Cohere also highlights, among the high-performing open models, DeepSeek V4.1 Flash, GLM 5.3 Flash, MiMo V2.6 Pro and Gemma 4.
There is, however, an important difference in the encrypted version: in Model Vault Encrypted, the third-party models available today are Gemma, GPT-OSS and Phi, plus Cohere's Embed, Rerank, Transcribe and Translate. DeepSeek, GLM, MiMo, Nemotron and Hy3 appear only in standard Model Vault, according to the table published by the company.
Where it is available
Model Vault is already available with regional serving in the United States, the European Union and Canada. Model Vault Encrypted is available in the United States, and its arrival in the European Union and Canada is presented as coming soon, with no date set in the announcement. Cohere also promises, in the coming months, more regions, larger models in the encrypted version and more models in each region.
The company's bet is so-called sovereign inference: the organization decides where data lives, who can access it and which model to use, instead of depending on a single provider or building and running its own inference cluster. Cohere cites RWS Language Weaver, which, according to the announcement, uses Model Vault to run models with zero data retention and control over where its clients' data goes.
Image credit: Cohere, on its official blog.
Sources
- Cohere — Leading Open AI Models Now Available in Model Vault
- Cohere Docs — Supported Models (Model Vault Encrypted)
- OkayNews — Cohere Adds Third-Party Models to Model Vault
Transparency: this article was prepared with the help of artificial intelligence, based on public sources on the internet, including Cohere's official announcement. The information was checked against the cited sources; consult the original sources to confirm the details.
Muse
