Vertex AI Model Garden

Vertex AI Model Garden Cheat Sheet

Model Garden is a centralized AI/ML model library on Vertex AI that helps you discover, test, customize, and deploy models and assets from Google and Google partners. It provides over 200 models, with consistent deployment patterns and built-in integration with Vertex AI’s tuning, evaluation, and serving capabilities.

Model Categories

Category	Description	Examples
Foundation models	Pretrained multitask large models that can be tuned or customized for specific tasks using Vertex AI Studio, API, or Python SDK	Gemini, Imagen, Veo, Chirp
Fine-tunable models	Models that you can fine-tune using custom notebooks or pipelines	Gemma, CodeGemma, PaliGemma
Task-specific solutions	Prebuilt models ready to use; many can be customized with your own data	Vision, Natural Language, Translation, Speech
Open models	Enterprise-ready open models from the community	Llama (Meta), Mistral AI, AI21, Falcon, BERT, T-5 FLAN, ViT, EfficientNet
Third-party models	Partner models available through Model Garden	Anthropic’s Claude family

Available Models by Type

Model Type	Models Included
Google First-Party	Gemini models (Gemini 3 Pro, Gemini 3 Flash, Gemini 3 with Nano Banana), Imagen (text-to-image), Veo (text-to-video and image-to-video), Chirp (speech-to-text)
Pre-trained APIs	Text-to-speech, Natural Language Processing, Translation, Vision
Open Source	Gemma, CodeGemma, PaliGemma, Meta’s Llama, Mistral AI, AI21, TII’s Falcon, BERT, T-5 FLAN, ViT, EfficientNet
Partner Models	Anthropic’s Claude Model Family

Key Advantages

Advantage	Description
Single location	All available models grouped in one place for easy discovery
Consistent deployment	Standard deployment pattern across different model types
Built-in integration	Seamless connection with Vertex AI tuning, evaluation, and serving
Managed deployment	Vertex AI handles model deployment and serving for generative AI models

Filtering Models

In the Google Cloud console, you can filter models by:

Filter	Options
Tasks	Select the task you want the model to perform
Model collections	Models managed by Google, partners, or you
Providers	Choose the model provider
Features	Select specific features you want in the model

Model Security Scanning

Model Source	Security Measures
Google-provided containers	Thorough testing, benchmarking, and active vulnerability scanning on serving and tuning containers
Third-party partner models	Model checkpoint scans to ensure authenticity
HuggingFace Hub models	Scanned by HuggingFace and third-party scanners for malware, pickle files, Keras Lambda layers, and secrets
Flagged models	Models deemed suspicious or capable of remote code execution are indicated in Model Garden but can still be deployed (review recommended)

Testing and Deployment Options

Method	Description	Charges
Playground	Quick text-based testing in model card with configurable parameters (temperature, output tokens)	Free (uses predeployed endpoints)
Spaces	Launch a working web application (Gradio app) deployed on Cloud Run that uses the model’s endpoint	Charged for machines used and Cloud Run instance
Prompt design	Test prompts in Vertex AI Studio with model parameters	Varies by usage

Model Garden Spaces

Spaces provides a way to launch sample applications for models like Gemini, Gemma, Llama, and Stable Diffusion.

Feature	Details
Availability	Only for projects not parented by an organization
Deployment	Deploys model in Vertex AI and sample app on Cloud Run
Access control	Options: Require authentication (via Identity Aware Proxy) or Allow public access
Security	Web application requires a secret key (appended to URL) for submitting prompts

IAM Permissions for Spaces

Action	Required Permissions	Purpose
Enable APIs	`serviceusage.services.enable`	Enable Cloud Run Admin, Artifact Registry, Cloud Build, Cloud Logging APIs
Grant permissions to service accounts	`resourcemanager.projects.setIamPolicy`	Grant Vertex AI Service Agent and Cloud Build Service Account roles
Deploy specific permissions	`storage.buckets.create`, `run.services.create`, `artifactregistry.repositories.create`, `run.services.setIamPolicy`	Upload source code, create Cloud Run service, create repository for container image, make service publicly accessible

Pricing

Activity	Pricing Model
Model tuning	Charged for compute resources at custom training rates
Model deployment	Charged for compute resources used to deploy model to endpoint (predictions pricing)
Colab Enterprise	See Colab Enterprise pricing
Spaces	Charged for machines used for deployment and Cloud Run instance hosting the app

New customers receive up to $300 in free credits to try Google Cloud AI and ML products.

Access Control

Set Model Garden organization policy at organization, folder, or project level to control access to specific models. For example, allow access to vetted models and deny access to all others.