Vertex AI SDK Deprecated: Migration Deadline June 24, 2026 — What Routing Teams Need to Know

The Vertex AI console disappeared from Google Cloud on May 21, 2026. If your team missed that news, here is the part that cannot wait: the Vertex AI SDK's generative AI modules are scheduled for removal on June 24, 2026 — 28 days from today. Any production code that imports vertexai.generative_models, vertexai.language_models, vertexai.vision_models, vertexai.tuning, or vertexai.caching will break on that date. The rebrand to Gemini Enterprise Agent Platform is cosmetic at the console level. The SDK removal is not.

What happened

Google completed two separate but related changes in May 2026:

Console migration (May 21). Vertex AI no longer appears as a top-level product in the Google Cloud Console. Searching for "Vertex AI" redirects to the new Gemini Enterprise Agent Platform. Model training, AutoML, Model Registry, and Endpoints are now sub-features under an agent-first hierarchy. The API endpoint remains aiplatform.googleapis.com — no URL changes required for existing HTTP clients — but the product branding and console navigation are fully replaced.

SDK deprecation clock (June 24 removal). The following modules in google-cloud-aiplatform were deprecated June 24, 2025 and will be removed on June 24, 2026:

• vertexai.generative_models
• vertexai.language_models
• vertexai.vision_models
• vertexai.tuning
• vertexai.caching

The replacement is the Google Gen AI SDK (google-genai package). The new SDK uses a unified vertexai.Client (or google.genai.Client) interface and supports both Gemini API and Vertex AI backends via the same library.

Additionally, Google has published an official migration guide from the OpenAI SDK to the Gen AI SDK — a path specifically for teams currently calling Gemini through an OpenAI-compatible proxy.

Why it matters for AI engineering teams

The console rename is annoying but harmless. The SDK removal is a breaking change with a fixed calendar date.

If your codebase imports vertexai.generative_models directly, it will throw an ImportError or AttributeError starting June 24. This affects:

• Internal wrappers that call Gemini models through the old Vertex AI SDK
• Fine-tuning pipelines using vertexai.tuning
• Vision or multimodal pipelines using vertexai.vision_models
• Any caching layer built on vertexai.caching

If your team calls Gemini through an OpenAI-compatible proxy (including routing gateways that translate /v1/chat/completions to Gemini), you are not affected by this particular SDK removal — the underlying HTTP API endpoint did not change. However, Google's migration guide actively promotes moving to the Gen AI SDK's native interface, which is worth evaluating for workloads where you want access to Gemini-specific features (thinking budgets, grounding, function calling with native tool schemas).

If you have SDK version pins in requirements.txt or pyproject.toml, check whether your pinned version of google-cloud-aiplatform still exposes the deprecated modules. A version pin alone will not protect you past June 24 if the upstream package removes the module at the package level. Verify your actual import paths.

The new Gemini Enterprise Agent Platform also ships two new features that affect routing and governance choices:

• MCP Server Registry (General Availability): The Agent Platform now has a managed MCP server registry, making it easier to attach tools to agents without custom scaffolding. This is relevant for teams evaluating whether to self-host MCP tool dispatch or hand it to Google's managed plane.
• Agent Identity: Each agent gets a cryptographic identifier for auditable action trails. If your governance or compliance workflow requires per-action attribution, this is now a first-class Google Cloud feature rather than something you have to build yourself.

The router/operator angle

The Vertex AI rename creates a subtle risk for routing teams: documentation and configuration strings scattered across code, runbooks, and SDK configs may still reference "Vertex AI" by name even when the underlying service now officially answers to "Gemini Enterprise Agent Platform."

Concretely, watch for these breakage vectors:

1. Import paths in Python wrappers. from vertexai.generative_models import GenerativeModel — this is the most common failure. Replace with import vertexai; client = vertexai.Client(project=..., location=...) and use the client.models.generate_content(...) interface from the new SDK.

2. SDK package names in CI/CD. If your Docker images or CI jobs pin google-cloud-aiplatform==1.x, the June 24 removal affects the package at source. Check whether a new google-cloud-aiplatform release removes these modules, or whether you should migrate to google-genai before the deadline.

3. Routing gateway model IDs. The Gemini model IDs themselves (gemini-3.5-flash-preview-05-20, etc.) are not changing — only the SDK wrapper layer. If your gateway routes requests to Gemini via its OpenAI-compatible interface, the model IDs and endpoint (https://generativelanguage.googleapis.com/v1beta/openai/) are stable.

4. Monitoring and telemetry labels. Dashboards, cost-allocation tags, and log filters that categorize traffic as "Vertex AI" calls may need to be relabeled to "Gemini Enterprise Agent Platform" or the underlying service name to stay aligned with Google's billing and IAM surfaces.

The practical migration decision is:

What TheRouter users should watch or try

If you route Gemini calls through TheRouter using an OpenAI-compatible interface, the June 24 SDK removal does not affect your gateway routing path. The HTTP API endpoint for Gemini is unchanged.

What to watch:

• Audit any internal Python code that wraps Gemini calls using the old vertexai.generative_models SDK. Even if the production routing path is gateway-based, test environments, eval scripts, or fine-tuning pipelines may still use the old SDK.

• Review your Gemini model IDs against the Agent Platform release notes. With the Vertex AI console gone, the canonical model catalog now lives under the Gemini Enterprise Agent Platform docs. Any model name that previously appeared in Vertex AI Model Garden may have a new documentation home.

• Evaluate the Gen AI SDK migration guide if you are building new integrations. The new SDK's unified client (vertexai.Client) gives you access to Gemini-specific capabilities — thinking budget control, native grounding, tool calling — that OpenAI-compatible proxies do not surface. For workloads where those capabilities matter, a hybrid approach (gateway for multi-provider routing and fallback; native SDK for Gemini-specific features) is worth modeling.

• Check billing labels. If you track Gemini cost separately from other providers in your ledger, verify that Google Cloud billing still uses the same service names and SKUs after the console migration. Product renames occasionally shift SKU labels.