Google Cloud Vertex AI

Google Cloud Vertex AI, now rebranded as the Gemini Enterprise Agent Platform, is Google's comprehensive platform for building, training, deploying, and governing both traditional machine learning models and modern AI agents at enterprise scale. Vertex AI was announced at Google I/O and launched generally available on May 18, 2021, unifying Google's previously separate AutoML and Cloud AI Platform offerings under a single API and interface, and later adding generative AI support with foundation models in June 2023. On April 22, 2026, at Google Cloud Next, Google announced the platform's evolution into the Gemini Enterprise Agent Platform, restructuring what was once a "model platform with agent features" into an agent-first architecture where model training, AutoML, and endpoints now sit as sub-features beneath a unified agent-building experience.

The platform gives developers access to over 200 Google and third-party AI models through Model Garden, including Google's Gemini 3.5 family, Anthropic's Claude models, and open models like Gemma, alongside tools for customizing models through fine-tuning and evaluation. Agent Studio offers a low-code environment for designing, testing, and managing prompts and agents using natural language, code, images, or video, while the Agent Development Kit (ADK) supports fully custom, code-first agent development. Newer capabilities include a managed agent runtime for scalable deployment, Memory Bank for persistent cross-session context, and integration with the Agent2Agent (A2A) protocol so agents built on different platforms can communicate.

Beyond agent building, the platform retains its full MLOps foundation, including Model Registry, Pipelines, Feature Store, and Model Evaluation, natively integrated with BigQuery and notebook environments like Colab Enterprise and Workbench, so data scientists and ML engineers can train, tune, and monitor models throughout their lifecycle. Google reports this shift is aimed at competing directly with fragmented multi-vendor AI stacks, offering a single platform spanning models, infrastructure, governance, and deployment. Notably, the underlying API endpoint remains unchanged, so existing integrations continue to work without requiring code changes despite the branding shift.

The platform uses pay-as-you-go pricing for the tools, storage, and compute resources used, with new customers getting up to $300 in free credits. Generative AI pricing starts at $0.0001 based on image input, character input, or custom training pricing for Imagen models, and text, chat, and code generation starts at $0.0001 per 1,000 characters based on input (prompt) and output (response). Custom model training pricing is based on machine type used per hour, region, and any accelerators used, available via a sales estimate or the pricing calculator. Notebooks are billed at the same rates as Compute Engine and Cloud Storage, plus separate management fees based on region, instances, and notebooks used. Pipelines start at $0.03 per pipeline run based on execution charges and resources used. Vector Search pricing is based on data size, queries per second (QPS), and number of nodes used. A pricing calculator and custom quotes from sales are available for detailed cost estimates.