What Is Vertex AI? Google Cloud’s AI Platform in 2026

Updated October 21, 2026. Vertex AI was Google Cloud’s platform for building, training, and deploying machine-learning models and AI applications. In 2026, Google evolved and renamed that platform Gemini Enterprise Agent Platform. Its capabilities continue under the new name, so current documentation and console labels use Agent Platform.

What is Vertex AI

What was Vertex AI?

Vertex AI brought Google Cloud’s machine-learning tools together in one managed environment. Teams could choose a ready-made model, train or customize a model, and build workflows to evaluate and serve it. It was a platform of services, not one AI model.

That distinction still helps when reading older tutorials: Gemini is a model family, while Vertex AI was the Google Cloud platform through which teams could access models and ML services.

What changed in 2026?

On April 22, 2026, Google introduced Gemini Enterprise Agent Platform as the evolution of Vertex AI. Google’s current documentation maps “Vertex AI Platform” to “Gemini Enterprise Agent Platform,” “Vertex AI Studio” to “Agent Studio,” and “Vertex AI API” to “Gemini Enterprise Agent Platform API.”

Google says future Vertex AI services and roadmap updates will be delivered through Agent Platform rather than as a separate standalone service. If an older guide points to a Vertex AI menu or feature name, check the official name-change table and follow the current product documentation.

For the announcement and its scope, see Google Cloud’s April 2026 launch post.

What can Gemini Enterprise Agent Platform do?

The platform supports model-based applications and AI agents across several stages:

Explore existing models. Model Garden helps teams discover, test, customize, and deploy models from Google and its partners. Model availability and deployment options vary, so review a model’s card and terms before choosing it. See the Model Garden guide.

Train or customize models. Teams can use managed training and other machine-learning services when a prebuilt model is not enough. Depending on the task, this may include AutoML, custom training, model evaluation, and repeatable pipelines.

Build agents and applications. Agent Studio offers a low-code visual environment. Developers can use the Agent Development Kit (ADK) to build code-first agents, then deploy supported agents to Agent Runtime.

Evaluate, govern, and operate. The platform includes tools for assessing model or agent quality, managing access, and observing deployed workloads. These tools help teams test behavior and manage production use; they do not remove the need for application-specific security and quality checks.

For the current product map and capabilities, start with Google’s Agent Platform overview.

How does the workflow usually look?

1. Define the task. Specify what the application should do, who will use it, what data it can access, and how you will judge a good result.

2. Pick an approach. Try an available model first. Add grounding or retrieval when responses need trusted source material. Consider tuning or custom training only when testing shows the simpler option does not meet your needs.

3. Evaluate with realistic examples. Check output quality, latency, safety, and failure cases using representative inputs before exposing the system to users.

4. Choose how to serve it. Online inference is for requests that need a response on demand. Batch inference is for larger jobs that can run asynchronously. The best choice depends on response-time needs, workload, and cost.

5. Monitor after launch. Set access controls, watch usage and errors, and re-evaluate when data, prompts, models, or user needs change.

How do you get started?

Choose or create a Google Cloud project, confirm billing is enabled, enable the Gemini Enterprise Agent Platform API, and arrange the required IAM permissions. Then select a Model Garden model or follow a current quickstart for your chosen workflow.

Google’s project setup guide covers the environment prerequisites. For a code-first agent example, use the ADK quickstart.

How much does it cost?

There is no single price for every workload. Costs can depend on the selected model and its usage, training or tuning, deployed compute, and related Google Cloud resources. Review the current Agent Platform pricing for the exact model or service, and estimate your expected usage before deployment. Prices and available options can change.

In short

If you searched for Vertex AI, you are looking for the predecessor name of Google Cloud’s current AI platform. Gemini Enterprise Agent Platform is the name to use for new documentation and workflows; the right service depends on whether you need a hosted model, a custom ML workflow, or an AI agent.