Vertex AI SDK Generative Module — when and how should I migrate?

Teams using deprecated Vertex AI SDK generative namespaces need a verified migration plan to the Google Gen AI SDK before those modules are removed, especially when code must span Gemini Developer API prototyping and Vertex AI enterprise authentication flows.

Adopt one Google Gen AI SDK codebase that switches between Gemini Developer API and Vertex AI, if you need shared code across prototyping and production.

Blockers

Who this is for

Candidates

Cut over fully to the Google Gen AI SDK on Vertex AI with ADC or service-account-backed auth

As of 2026-03-15, this is the default production migration path for teams already committed to Google Cloud. Google documents that the Generative AI module in the Vertex AI SDK was deprecated on June 24, 2025, will be removed on June 24, 2026, and that the Google Gen AI SDK has full feature parity with the deprecated modules and packages while adding additional capabilities.

When to choose

Best for enterprise + compliance or microservices + monorepo environments where production traffic should stay on Vertex AI, IAM-based auth matters more than API-key convenience, and you want one supported SDK before the June 24, 2026 shutdown. Prefer this when your code currently depends on deprecated Python, JavaScript, Java, or Go generative namespaces from the Vertex AI SDK.

Tradeoffs

You align with Google's replacement SDK, keep Vertex AI enterprise controls, and can authenticate with ADC or a Google Cloud API key bound to a service account. The tradeoff is migration work across dependencies, imports, client initialization, and method shapes, plus the normal Google Cloud setup for billing, IAM, and project configuration.

Cautions

AI Studio API keys are not supported in Vertex AI. As of 2026-03-15, the Vertex AI quickstart says existing Google Cloud users should use ADC or an API key bound to a service account, and the migration docs say supported regions can differ from the Gemini Developer API. Recheck the live Vertex AI pricing page before rollout if approval depends on exact token, grounding, or batch prices.

Sources

Adopt one Google Gen AI SDK codebase that switches between Gemini Developer API and Vertex AI

As of 2026-03-15, Google positions the Google Gen AI SDK as a unified interface across the Gemini Developer API and the Gemini API on Vertex AI. Google says that, with a few exceptions, code that runs on one platform will run on both, making this the cleanest option when one codebase spans prototype API-key flows and enterprise Vertex AI deployments.

When to choose

Best for small-team + low-ops or monorepo + microservices teams that need shared model-calling code across local prototyping, staging, and production, but still want the option to move workloads from Gemini Developer API to Vertex AI without a full rewrite. Prefer this when the real constraint combination is Gemini API for fast iteration plus Vertex AI for enterprise auth, regional controls, or SLA-backed production.

Tradeoffs

One SDK reduces duplicated wrappers and lets you switch by configuration: API key for Gemini Developer API, or project and location with Vertex AI settings for Vertex AI. The tradeoff is that backend-specific features still exist, so a shared abstraction has to gate or branch around services that are only available on one backend.

Cautions

Google explicitly notes that some services are only available in a specific backend and that unsupported calls can raise `UnsupportedOperationException`. If you standardize on one codebase, keep backend-specific integration tests and avoid assuming Gemini Developer API auth, regions, quotas, or enterprise controls match Vertex AI.

Sources

Use Vertex AI express mode as a short-lived API-key bridge, then upgrade to full Vertex AI

As of 2026-03-15, Vertex AI express mode lets teams start with an API key and the Google Gen AI SDK while staying on the Vertex AI API surface instead of the Gemini Developer API. Google documents a simplified setup, a 90-day free tier for new Google Cloud users, and an upgrade path to full Google Cloud without service interruption.

When to choose

Best for small-team + low-ops or serverless + cost-sensitive teams that need to stop depending on deprecated Vertex AI SDK namespaces quickly, still want Vertex AI endpoints, and are not ready to do full Google Cloud project and IAM rollout on day one. This fits staged migrations where prototypes or internal tools can tolerate express-mode limits while the production path is being moved to standard Vertex AI.

Tradeoffs

You get a faster API-key-based on-ramp to the replacement SDK and can later upgrade to full Google Cloud, carrying workloads, saved prompts, and API keys forward. The tradeoff is that express mode exposes only a subset of Vertex AI features, simplifies away organizations and projects, and is not the same operational model as full Vertex AI.

Cautions

Express mode is Preview and subject to Pre-GA terms. As of 2026-03-15, Google documents no SLA in the free tier, a `@gmail.com` requirement to sign up, 90-day access unless billing is enabled, and data marked for deletion 30 days after the free period ends if billing is never turned on. Use it as a bridge, not as the long-term enterprise target.

Sources

Facts updated: 2026-03-15
Published: 2026-04-03

Related questions

Try with your AI agent

$ npm install -g pocketlantern
$ pocketlantern init
# Restart Claude Code, Cursor, or your MCP client, then ask:
# "Vertex AI SDK Generative Module — when and how should I migrate?"
Missing something? Request coverage