Gemini 3.5, Agent Builder, and the platform Google just rebranded from Vertex AI — model catalog, pricing, architecture, and a working technical guide for migrating off Google AI Studio.
Data verified July 2026 · sourced from Google Cloud's official documentation
At Google Cloud Next 2026, Vertex AI was rebranded to the Gemini Enterprise Agent Platform and consolidated with Agentspace. If you already use Vertex AI, nothing breaks — it's the same platform under a new, more agent-focused name.
Gemini Enterprise is Google Cloud's governed, project-based platform for building with Gemini and 200+ other models — as opposed to Google AI Studio, which is a free, API-key-based sandbox meant for prototyping. Gemini Enterprise wraps model access in IAM, VPC Service Controls, and GCP billing, and bundles the Agent Development Kit (ADK), a managed Agent Engine runtime, and Vertex AI Search into one platform for production agents.
From the flagship Gemini 3.5 Pro to the cost-optimized Flash tier and the wider Model Garden.
Deep multimodal reasoning across text, image, video, and audio in a single call, for the most complex enterprise workloads.
High-volume production traffic — launched May 2026 at roughly 25% below the previous Gemini 3.1 Pro price while beating it on coding and agentic benchmarks.
Still supported for workloads not yet migrated to the 3.5 family; being phased out as 3.5 Flash undercuts it on both price and benchmarks.
Running Anthropic Claude or open-weight Llama/Gemma models inside the same governed Google Cloud project as Gemini — over 200 models available.
Enterprise-grade image generation for marketing and creative workflows, governed under the same IAM policies as your text models.
Input token pricing — Flash launched roughly 25% cheaper than the previous-generation Pro while beating it on coding and agentic benchmarks.
USD per 1 million input tokens
Six building blocks bundled into one pay-as-you-go platform.
A code-first framework for building agents in Python or Java, with full control over tools, memory, and orchestration logic.
A low-code, visual builder for teams that want to assemble agents without writing orchestration code from scratch.
The managed runtime that hosts and scales agents in production — billed at roughly $0.0864 per vCPU-hour and $0.0090 per GB-hour of memory.
Enterprise search and retrieval grounding for RAG, priced at $1.50-$6.00 per 1,000 queries depending on tier.
Session and long-term memory storage for agents, billed at $0.25 per 1,000 events — lets agents recall context across sessions.
A single meta-endpoint for Gemini requests — set a cost, quality, or balanced preference and it automatically routes to the right model tier.
Fine-grained identity and access management plus VPC Service Controls to create a security perimeter around your project data.
Region and multi-region processing options so enterprise data stays within required geographic boundaries.
Centralized logging, usage monitoring, and governance controls across every agent and model call in a project.
New customers can get up to $300 in free credits to evaluate the Agent Platform and other Google Cloud products before committing spend.
The path production workloads take once free-tier quotas stop being enough.
Set up (or reuse) a Google Cloud project and enable the Vertex AI / Gemini Enterprise API — this replaces the standalone API key from Google AI Studio with project-based access.
Project setup, API enablement, billing account linkage
Replace the google-generativeai (AI Studio) package with the unified Google Gen AI SDK — the Vertex AI SDK itself is being retired for Gemini after June 2026, so the Gen AI SDK is the forward-compatible path.
pip uninstall google-generativeai; install google-genai
Remove the plaintext GOOGLE_API_KEY environment variable and authenticate with a service account or Application Default Credentials (ADC) instead — the standard IAM-based pattern for production GCP workloads.
Service account creation, ADC configuration, Workload Identity for GKE
Point requests at your project and region instead of the public Gemini API endpoint, and update the model identifier to the current Gemini 3.5 generation.
Region selection, model ID updates, endpoint configuration
Once traffic is billed through GCP directly, you're no longer bound by Google AI Studio's free-tier quota or spending cap — validate rate limits and scale up gradually.
Quota request, production traffic ramp-up, cost monitoring
Gemini models become available on Vertex AI alongside PaLM, giving enterprises project-based, IAM-governed access.
Vertex AI Agent Builder adds the ADK, Agent Engine runtime, and Vertex AI Search as core building blocks for production agents.
Vertex AI is rebranded and consolidated with Agentspace into the Gemini Enterprise Agent Platform. Existing customers keep the same services under the new name — no forced migration for the platform itself.
New Gemini capabilities ship only in the unified Google Gen AI SDK; the older Vertex AI SDK stops receiving Gemini updates, pushing every remaining integration toward the new SDK.
Launches at roughly 25% below Gemini 3.1 Pro pricing while outperforming it on coding and agentic benchmarks.
Vertex AI Search grounds Gemini responses in your own documents, wikis, and databases with citations.
ADK-built agents that hand off tasks to each other — research, drafting, and review chained into one automated workflow.
Gemini models called directly against BigQuery data for natural-language analytics without moving data out of Google Cloud.
Gemini plus Imagen for combined text-and-image generation workflows in marketing and product teams.
The full three-way comparison — model breadth, pricing model, and governance — plus an interactive model explorer.
Compare all three →GPT-5, Azure AI Foundry, and Provisioned Throughput pricing, explained in the same depth.
Read the Azure OpenAI guide →We'll assess your workloads, design your Agent Builder architecture, and handle the migration from Google AI Studio or another provider.