Unlock the Power of AI with Google Vertex AI

Vertex AI is Google Cloud's machine learning and generative AI platform. On April 22, 2026, Google renamed it Gemini Enterprise Agent Platform (product pages also say "Gemini platform" or "Agent Platform"), and old cloud.google.com/vertex-ai links now redirect there. Google's release notes map the old names: Vertex AI Studio is now Agent Studio, Vertex AI Agent Engine is Agent Runtime, Vertex AI Search is Agent Search, the Vertex AI API is the Agent Platform API, and Model Garden keeps its name. Google's pricing page says costs remain the same as for the legacy products the platform supersedes.

You use it inside a Google Cloud project. Agent Studio is a console for designing and testing prompts against Gemini models; Model Garden offers 200+ Google, third-party and open models, including Gemini 4 Argon, Anthropic's Claude family and Gemma; the Agent Development Kit (ADK) and Agent Runtime are for building and hosting agents; and the classic ML side covers custom training, AutoML, Pipelines, Model Registry, Feature Store, model monitoring and Colab Enterprise or Workbench notebooks connected to BigQuery. Access is through the console, REST API and SDKs for Python, JavaScript, Java and Go.

There is no subscription: you pay for model tokens, compute and storage you use, billed to a Google Cloud account. New Google Cloud customers get $300 in free credits. Only requests that return a 200 response are charged.

Selected Gemini model prices on the Global endpoint (standard tier, per 1 million tokens, prompts up to 200K tokens):

Non-global (regional) endpoints cost about 10% more (for example $2.20 / $11.00 for Gemini 4 Argon during the promotion). The Gemini 4 page also lists a Priority tier ($3.60 / $18.00 promotional) and a Batch tier ($1.00 / $5.00 promotional). Worked example: 1,000 requests of 2,000 input and 500 output tokens on Gemini 3.8 Flash at the introductory rate costs 2M x $0.75 + 0.5M x $3.75 = $1.50 + $1.875, about $3.38.

Other published rates:

• Imagen 4: $0.04 per image (Fast $0.02, Ultra $0.06).
• Grounding with Google Search or Google Maps: 5,000 queries a month free across Gemini 3 and 4 models, then $14 per 1,000. Grounding with your own data: $2.50 per 1,000 prompts.
• Agent Runtime (formerly Agent Engine): $0.085 per vCPU-hour and $0.009 per GiB-hour of memory, rounded to the second; idle time between turns is not billed. Sessions storage is $0.30 per GiB-month.
• Pipelines: $0.03 per pipeline run plus the resources used.
• Custom training and notebooks: billed by machine type, accelerator and region at Compute Engine rates; Google points to its calculator or sales for estimates.

Google notes that from November 13, 2026, Gemini 4 completions longer than 32K tokens are charged for retrievable tokens even if the request then times out.

Prices checked on the vendor's pricing page on November 2, 2026. See Generative AI pricing and Agent Platform pricing.

What Google's own pages list for the platform:

Vertex AI (now Agent Platform) suits teams already on Google Cloud who want Gemini and other models behind Google Cloud's identity, regional controls, billing and data services such as BigQuery, and ML teams who want training, deployment and monitoring in one place. It is a developer and data-science platform, not a ready-made chat app, and the console and billing model take time to learn.

If you only need to call Gemini from code and do not need Google Cloud's enterprise controls, the Gemini API through Google AI Studio is a simpler start. Teams standardised on AWS should compare Amazon Bedrock, and Microsoft shops Azure AI Agent Service. For open models and community hosting, see Hugging Face.

On PeerSpot, Google Vertex AI averages 4.1 out of 5 from 15 reviews. Reviewers there praise integration with other Google Cloud services, the strength of Gemini's language and vision models, scalability and having build, deploy and monitor in one place. Recurring complaints are that the platform is complex to navigate and to switch services on, split views on whether pricing is affordable, mixed support response times and slow AutoML jobs. Coverage of the April 2026 rename in THE Journal relays Google's description of the platform as "an evolution of Vertex AI" built out around agents; most PeerSpot reviews predate the rename, and older tutorials use the previous names.