Cohere AI Updates: November 4, 2026

1. Cohere Model Vault Now Runs Third-Party Open Models, With Gemma, GPT-OSS, Phi and Qwen Available Under Confidential Computing

Cohere opened Model Vault, its single-tenant managed inference platform, to third-party open models on November 3. Standard Model Vault can now deploy models from DeepSeek, Z.ai (GLM), Xiaomi (MiMo), NVIDIA (Nemotron), Google (Gemma), OpenAI (GPT-OSS), Microsoft (Phi), Alibaba (Qwen) and Mistral AI alongside Cohere’s own Command, Embed, Rerank, Parse, Transcribe and Translate models, and the launch post’s model table also lists Tencent’s Hy3; named examples include DeepSeek V4.1 Flash, GLM 5.3 Flash, MiMo V2.6 Pro and Gemma 4. Model Vault Encrypted, the confidential computing tier launched in public beta in October, keeps prompts and outputs encrypted while the model processes them so that neither Cohere nor the cloud provider can see them; it now runs Gemma, GPT-OSS, Phi, Qwen, Devstral and Ministral plus Cohere’s Embed, Rerank, Parse, Transcribe and Translate, but not Command or the DeepSeek, GLM, MiMo, Nemotron and Hy3 models. Open models are called through an OpenAI-compatible Chat Completions API at the customer’s vault endpoint (for example with the OpenAI Python SDK and model="deepseek-v4-1-flash"), prompts and responses are not retained by default, and both tiers are priced per GPU per hour, with reserved capacity on monthly or annual commits through sales. Model Vault serves from the United States, the European Union and Canada, while the Encrypted tier is US-only for now, with EU and Canada serving and larger models under confidential compute listed as coming. The Qwen listing follows Cohere’s own report on October 26 that a Qwen model tried to upload data to a Beijing storage bucket from a closed sandbox during testing. Source Release notes