Google Vertex AI

Vertex AI is Google Cloud's managed AI platform. In April 2026 Google renamed it Gemini Enterprise Agent Platform. Enterprise developers and ML teams use it to pick from more than 200 models, build agents in low-code or code-first tools, and run and govern those agents in production. It replaces the separate model, agent and MLOps tools a team would otherwise have to wire together.

Gemini Enterprise Agent Platform is Google Cloud's managed platform to build, scale, govern and optimize AI agents. It is the direct successor to Vertex AI. Google announced Vertex AI on May 18, 2021 at Google I/O as a unified MLOps platform and claimed it needed nearly 80% fewer lines of code to train a model than competing platforms. The launch bundled Vertex Pipelines, Feature Store, Experiments and Model Monitoring. On April 22, 2026, at Google Cloud Next, Google renamed the product. It said all Vertex AI services and roadmap evolutions would from then on ship only through the Agent Platform, not as a standalone service. Model Garden gives access to more than 200 models, including Gemini 3.1 Pro, Gemma 4 open models, Lyria 3 and Anthropic's Claude Opus, Sonnet and Haiku. Teams build agents in the low-code Agent Studio or the code-first Agent Development Kit (ADK). Google says six trillion Gemini tokens a month go through ADK. Agents run on Agent Runtime, which offers sub-second cold starts and supports agents that run for days, with Agent Sandbox and Agent Memory Bank alongside. Governance comes from Agent Identity, Agent Registry, Agent Gateway with Model Armor protections, and anomaly and threat detection surfaced through Security Command Center. Agent Simulation, Evaluation, Observability and Optimizer cover quality. Google names Comcast, L'Oreal, PayPal, Payhawk, Geotab and Color Health as customers. The platform is billed pay-as-you-go by token, compute hour and service meter.

Category

Agent Development

Pricing

Usage-based

Target Market

CTOs, CIOs, Enterprise Developers, Data Scientists, ML Engineers, Platform Engineering Teams

Deployment

Cloud-only, API-based

Founded

2021

Headquarters

Mountain View, United States

Team Size

500+

• ✓
  Model Garden

  First-class access to more than 200 models, including Gemini 3.1 Pro, Gemma 4 and Anthropic Claude, so teams can choose the best model per task.

• ✓
  Agent Studio and Agent Development Kit

  Low-code visual building in Agent Studio. Logic can be exported into the code-first ADK, so prototypes carry into full engineering without a rewrite.

• ✓
  Agent Runtime, Sandbox and Memory Bank

  Managed hosting with sub-second cold starts and support for agents that run for days. A hardened sandbox runs model-generated code, and Memory Bank keeps curated long-term memory.

• ✓
  Agent Identity, Registry and Gateway

  Each agent gets a cryptographic ID, is indexed in a central registry, and goes through a gateway that enforces security policy and Model Armor protections.

• ✓
  Agent Threat and Anomaly Detection

  Flags unusual agent reasoning and malicious activity, such as reverse shells, and surfaces it in a Security Command Center-powered Agent Security dashboard.

• ✓
  Agent Simulation, Evaluation and Optimizer

  Tests agents against synthetic users and scores live traffic with multi-turn autoraters. It also clusters real failures and suggests better system instructions.

• ✓
  MLOps heritage tooling

  Keeps Vertex AI's training, tuning, pipelines, feature store and endpoint hosting, so predictive ML and generative agents share one platform.

• •
  Customer-service agent rebuild

  Comcast is rebuilding its Xfinity Assistant on ADK to raise digital containment and route fewer conversations to human agents.

• •
  Finance back-office automation

  Payhawk built a Financial Controller Agent on the platform and reports cutting expense-submission time by more than 50%.

• •
  Healthcare virtual care assistant

  Color Health runs a Virtual Cancer Clinic assistant on a platform covered by Google Cloud's HIPAA Business Associate Agreement.

• •
  Enterprise knowledge agent

  Burns & McDonnell is building a knowledge agent with ADK so employees can query internal engineering and project information in natural language.

• •
  Agentic commerce and payments

  PayPal uses ADK with the Agent Payment Protocol (AP2) to build agents that can complete trusted payment transactions for users.

There is no subscription. Billing is pay-as-you-go across many meters: input and output tokens per model, agent runtime vCPU-hours and memory GB-hours, sessions, search, GPU/TPU node-hours for training and endpoints, and data services. List prices are published. According to CloudZero (September 2026), thinking tokens bill as output, idle endpoints keep billing, and prices jump above 200K context. Current Flash introductory rates double on January 1, 2027.

This page was written from 7 sources, 4 on domains other than cloud.google.com.

1. 1.cloud.google.com — introducing gemini enterprise agent platformvendor
2. 2.cloud.google.com — google cloud launches vertex ai unified platform for mlopsvendor
3. 3.cloud.google.com — hipaavendor
4. 4.peerspot.com — google vertex ai reviews
5. 5.cloudzero.com — google vertex ai pricing
6. 6.hn.algolia.com — search
7. 7.virtualizationreview.com — google cloud next 26 gemini enterprise agent platform leads