🇺🇸 USA · Google Vertex AI

Status: 🟩 COMPLETE 🟦 LIVING Last updated: 2026-06-26 Plain-English tagline: Google Cloud’s managed AI platform. The enterprise way to use Gemini, Imagen, Veo, Lyria + open-weight models like Claude, Llama, Mistral. With AUS data residency (Sydney + Melbourne).

Front-matter facts

Field	Value
Vendor	Google Cloud (Mountain View, USA)
Country / origin	🇺🇸 USA
Recommended for Australian users?	✅ Yes — Vertex AI in Sydney (`australia-southeast1`) + Melbourne (`australia-southeast2`) for AUS data residency
Privacy summary	No training on customer data; tenant-isolated; AUS data residency via Sydney / Melbourne; IRAP Protected available
Free tier	Some free quota via Google Cloud Free Program + service-specific free tiers
Paid tiers	Pay-per-token + provisioned-throughput; AUD via Google Cloud Australia
First released	May 2021 (replaced AI Platform); evolved continuously
Last reviewed	2026-06-26
Official site	https://cloud.google.com/vertex-ai

What it is

Vertex AI is Google Cloud’s managed AI platform — the enterprise-grade way to access Google’s AI capabilities and broader model catalog. Where Google AI Studio is for prototyping (free, simpler, but trains on free-tier inputs), Vertex AI is for production (paid, fully enterprise, no-train).

Vertex AI provides:

Google’s first-party models

Gemini (Pro / Flash / Deep Think variants)
Imagen (Imagen 4 for image generation)
Veo (Veo 3 for video generation)
Lyria (music generation)
Embeddings (text and multimodal)
PaLM (legacy)

Third-party models (Model Garden)

Anthropic Claude (full lineup; Anthropic-Google partnership)
Meta Llama (4 / 5 + variants)
Mistral models
Cohere models
Many open-weight models from Hugging Face

Platform features

Vertex AI Studio — UI for prompt engineering
Vertex AI Workbench — managed Jupyter for ML work
Vertex AI Pipelines — ML workflow orchestration
Vertex AI Search — RAG-as-a-service (Google’s enterprise search)
Vertex AI Agents (Agent Builder) — agentic workflows
AutoML — train custom models without code
Custom training — full-control ML training on TPUs / GPUs
Model Monitoring — drift detection, performance tracking

What you’d use it for

Production Gemini API with enterprise terms (vs Google AI Studio for prototyping)
AUS data residency for Gemini specifically (Sydney + Melbourne)
Multi-model strategy via Model Garden — Gemini + Claude + Llama from one platform
Enterprise RAG via Vertex AI Search
Custom ML training at scale (TPU access)
AI workflows / agents via Agent Builder
Regulated AUS workloads needing Google Cloud + AI
Google Workspace customers for ecosystem integration

When NOT to use Vertex AI:

For consumer Gemini chat (use gemini.google.com)
For prototyping (use Google AI Studio — simpler, free)
For non-AI cloud workloads (use other GCP services)

How to use from Australia

Google Cloud account with billing enabled
Create / select a project in australia-southeast1 (Sydney) or australia-southeast2 (Melbourne)
Enable Vertex AI API
Browse Model Garden for available models
Use Vertex AI Studio for testing, or call via API

Example API call (Python)

import vertexai
from vertexai.generative_models import GenerativeModel
 
vertexai.init(project="your-project", location="australia-southeast1")
model = GenerativeModel("gemini-3-pro")
response = model.generate_content("Hello")

What it costs

Pay-per-token (Gemini and most models)

Gemini 3 Pro: ~US $1.25/$ 5 per million input/output tokens (verify current)
Gemini 3 Flash: ~US $0.15/$ 0.30 per million
Claude (via Vertex): per Anthropic Bedrock-comparable rates
Imagen / Veo / Lyria: per-image / per-second / per-track pricing

Provisioned throughput

Reserved capacity for guaranteed latency
Monthly commitment

Free quotas

Google Cloud Free Program (USD $300, 90 days)
Vertex AI specific free quotas vary by model

Hidden costs

Vertex AI Search uses underlying Cloud Storage + indexing — additional infrastructure costs
Custom training on TPUs can be expensive — use spot / preemptible for cost savings

How it compares to alternatives

Aspect	Vertex AI	AWS Bedrock	Azure OpenAI Service / AI Foundry	Google AI Studio
Native models	Gemini + Imagen + Veo + Lyria	Amazon Nova	OpenAI GPT + MAI	Gemini
Third-party model catalog	Claude + Llama + Mistral	Claude + Llama + Mistral + Cohere	OpenAI + MAI + Llama + Mistral	Limited
AUS data residency	Sydney + Melbourne	Sydney	Australia East	Limited
Enterprise tier	Production-grade	Production-grade	Production-grade	Prototyping-grade
No-training commit	Yes	Yes	Yes	NO (free tier trains)
Pricing	Per-token + PTU	Per-token + PT	Per-token + PTU	Free / cheap
Best for	Google models + GCP shops	AWS shops + Anthropic	Microsoft shops + OpenAI	Prototyping

For AUS organisations wanting Gemini at production scale with data residency, Vertex AI is the obvious pick.

Privacy / data handling

No training on customer data — contractually committed
AUS data residency via Sydney + Melbourne
IRAP Protected available
HIPAA, GDPR, SOC 2, ISO 27001, FedRAMP compliant
Tenant-isolated by Google Cloud project
Customer-managed encryption keys (CMEK) via Cloud KMS
Audit logs via Cloud Audit Logs

Recent changes

2026: Gemini 3 family in Vertex; Imagen 4 + Veo 3 broadly available
2025: Model Garden expanded (more open-weight models)
2024: Agent Builder launched
2021: Initial Vertex AI launch (replacing AI Platform)

Gotchas

Model availability per region varies — verify your specific models in Sydney / Melbourne
Pricing across providers in Model Garden can be opaque — read each model’s pricing carefully
Vertex AI vs Google AI Studio confusion is common — AI Studio for prototyping (free, with training caveat), Vertex AI for production
Quota management is real — request uplift for production volumes
Anthropic Claude via Vertex uses Anthropic’s commercial terms within Google Cloud billing — same models, slightly different access pattern
For very small projects, Vertex AI is heavier than needed; Google AI Studio simpler
AUS billing tax-invoice: ensure billing account configured for Australia for proper GST handling

Tech & AI, Explained

Explorer

vertex-ai