Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- Google Cloud's November 2, 2026 release notes mark serving tuned Gemini 3.x models on global endpoints as generally available (GA).
- The documented sequence is to copy an existing tuned model to global, create a global endpoint, and deploy the copied model to it.
- Global does not support customer-managed encryption keys (CMEK) or data residency zones (DRZ); tuned model adapters for global models are stored in the US region.
- Global inference uses aiplatform.googleapis.com. The tuning job instructions for Gemini 3 and later still specify the us or eu multi-region endpoint.
- Check the exact model, project permissions and current charges separately. A general tuning support list does not establish eligibility for every global configuration.
What is now generally available?
Google Cloud's November 2, 2026 Gemini Enterprise Agent Platform release notes announce general availability for serving tuned Gemini 3.x models on global endpoints. The announcement describes copying a model through the Model Registry models:copy API and deploying it to a global serving endpoint.
This update matters to teams choosing a serving endpoint for an existing supported tuned model. The announcement is not a project-specific eligibility list: the sources reviewed here do not establish that every model version, account or project meets the same global deployment conditions. They also do not promise a particular change in speed, quality or cost after migration.
First, does global meet your encryption and residency requirements?
The official guide explicitly states that the global region supports neither Customer-Managed Encryption Keys (CMEK), which let customers manage encryption keys, nor Data Residency Zones (DRZ). If your project requires either capability, establish an applicable configuration before choosing global.
The guide also says that tuned model artifacts, or adapters, for global models are stored in the US region because Cloud Storage has no global bucket location. An adapter is an artifact produced by tuning. This statement does not provide a complete map of where training data, requests, logs and every processing stage are stored or handled, so it cannot establish the residency of an entire application.
How do you copy and deploy an existing tuned model?
First, identify the existing model and its source location. The official example copies a model from a source such as us or eu to locations/global/models:copy in the destination project. The sourceModel value must identify the actual source project, location and model ID. This operation copies an existing tuned model.
Second, create a global endpoint under locations/global/endpoints in the destination project and obtain its endpoint ID. Third, call deployModel on that endpoint, pointing the deployment configuration at the copied model under locations/global/models. Copying the model, creating the endpoint and deploying the model are three operations. A completed copy alone does not establish that the endpoint can accept inference requests.
Which addresses do inference and tuning jobs use?
For global serving, the documented inference address format is https://aiplatform.googleapis.com/v1/projects/PROJECT_ID/locations/global/endpoints/ENDPOINT_ID:generateContent. The hostname is aiplatform.googleapis.com, and the path uses global and the actual endpoint ID. Check your client configuration against the global instructions rather than carrying over the multi-region format that uses .rep.googleapis.com.
The serving location and the tuning job location are separate settings. For Gemini 3 and later, the guide instructs you to use the us or eu multi-region endpoint for tuning jobs. Copying an existing model to global does not also move the original tuning job to global. No model copy, deployment or inference call was executed for this article.
What else should you check before deployment?
Check tuning support and global serving conditions for the exact base model and version. The guide's general tuning list spans Gemini 3.x and 2.5 models; it is not a model-by-model eligibility table for this Gemini 3.x global announcement. Console features have version limits too: the guide limits certain console creation, testing and metrics-viewing operations to Gemini 2.5 and earlier. Do not assume those console paths apply to 3.x.
Next, check access to the source model and destination project, the IAM permissions of the identity performing each operation, and the current charges applicable to copying, deployment and inference. The announcement and steps reviewed here do not establish a complete least-privilege IAM table for every operation, current prices or account-specific eligibility. Confirm those details in the current official documentation before acting; GA does not mean free use. Finally, compare the US adapter storage and the CMEK and DRZ limits with your project's requirements.
Sources
Frequently Asked Questions
Will existing models switch to global automatically?
The announcement and guide reviewed here do not describe an automatic migration. They describe explicitly copying an existing model, creating a global endpoint and deploying the model. The announcement alone does not establish that an existing endpoint has changed.
Can I change the hostname and keep my original endpoint ID?
The documented flow creates a global endpoint and deploys the model copied to global onto it. Requests should use the actual global endpoint ID in the destination project. The guide does not guarantee that an endpoint ID from the original region can be reused directly in the new path.
Is global always faster or cheaper than us or eu?
The reviewed sources provide neither a performance comparison nor a complete price comparison that establishes that conclusion. Assess the same task and billing basis after confirming model eligibility, project permissions and data requirements.
Was this capability released on November 3?
This article uses the November 2, 2026 entry date in the release notes. The guide's November 3, 2026 UTC footer is a documentation update date. It does not establish the feature's first launch time or when an individual account received access.