Differences in Response Models between the Vertex AI SDK and the Gen AI SDK

When migrating a Python-based AI application from the Vertex AI SDK to the Gen AI SDK, I made an interesting discovery: the Gen AI SDK uses a Pydantic-based response model (GenerateContentResponse), which means you can serialize it with model_dump() or model_dump_json().

For anyone unfamiliar with the current landscape, it can be confusing that Google offers multiple official SDKs for working with the Gemini API. Below is some background before we dive in. At the moment, Gemini exposes two main APIs and three Python SDKs.

---

Gemini APIs

1. Gemini API in Vertex AI

• Access Gemini models via Google Cloud’s Vertex AI
• Requires a Google Cloud project
• IAM-based authentication and access control
• Per-project quotas that throttle usage as needed

See: “Migrate from the Gemini Developer API to the Vertex AI Gemini API” (Google Cloud Docs)

2. Gemini Developer API

• Access Gemini through Google AI Studio
• Works even without a Google Cloud project
• Generous free tier—ideal for learning and prototyping
• Enterprise-grade features and advanced settings are limited

See: “Get a Gemini API key | Google AI for Developers”

---

Official Python SDKs

Google currently maintains three Python SDKs:

1. Google Gen AI SDK (google-genai)

• Supports both the Gemini Developer API and the Vertex AI Gemini API
• A single code base that handles API-key auth (AI Studio) and IAM auth (Vertex AI)
• Newer than the others and updated most frequently

Docs: https://googleapis.github.io/python-genai/

2. Vertex AI SDK (google-cloud-aiplatform)

• Dedicated to the Gemini API in Vertex AI
• Lets you use Gemini models through Vertex AI

Repo: https://github.com/googleapis/python-aiplatform

3. Google AI Python SDK (google-generativeai)

• Targets the Gemini Developer API only
• Does not work with Vertex AI
• Now deprecated; scheduled for EOL at the end of August 2025

Repo: https://github.com/google-gemini/deprecated-generative-ai-python

---

How the Response Models Differ

Both the Vertex AI SDK and the Gen AI SDK can call the Gemini API in Vertex AI, but their usage patterns differ slightly. Below are minimal examples.

Example with the Vertex AI SDK

from vertexai.generative_models import GenerativeModel
import vertexai

PROJECT_ID = "*****************"
REGION = "us-central1"

vertexai.init(project=PROJECT_ID, location=REGION)

model = GenerativeModel("gemini-2.0-flash")

prompt = "What is the capital of Japan?"
response = model.generate_content(prompt)

print(response.text)           # -> Tokyo is the capital of Japan.

Example with the Gen AI SDK

from google import genai
from google.genai.types import HttpOptions

client = genai.Client(
    vertexai=True,
    project="*****************",
    location="us-central1",
    http_options=HttpOptions(api_version="v1"),
)

response = client.models.generate_content(
    model="gemini-2.0-flash",
    contents="What is the capital of Japan?",
)

print(response.text)           # -> Tokyo is the capital of Japan.

Although you still retrieve the generated text via response.text, the underlying response object differs:

SDK	Response class
Vertex AI SDK	GenerationResponse
Gen AI SDK	GenerateContentResponse

Each response bundles the generated content plus rich metadata. If you want to dump everything to JSON—for example, to inspect intermediate artifacts—here’s how you do it with each SDK.

Serializing with the Vertex AI SDK

import json

import vertexai
from vertexai.generative_models import GenerativeModel

vertexai.init(project="*****************", location="us-central1")

model = GenerativeModel("gemini-2.0-flash")
response = model.generate_content("What is the capital of Japan?")

print(json.dumps(response.to_dict(), indent=2, ensure_ascii=False))

GenerationResponse exposes a handy to_dict() method.

Serializing with the Gen AI SDK

GenerateContentResponse is built on Pydantic’s BaseModel, so you use model_dump() (or model_dump_json()) instead:

from google import genai
from google.genai.types import HttpOptions
import json

client = genai.Client(
    vertexai=True,
    project="*****************",
    location="us-central1",
    http_options=HttpOptions(api_version="v1"),
)

response = client.models.generate_content(
    model="gemini-2.0-flash",
    contents="What is the capital of Japan?",
)

print(json.dumps(response.model_dump(mode='json'), indent=2, ensure_ascii=False))

Tip
If you call to_dict() on a GenerateContentResponse, you’ll get an error—one of those “gotchas” to watch for when migrating.

Google’s migration guide and GitHub issues both mention this explicitly:

• Migration guide: Generate-content section (ai.google.dev)
• Issue tracker: googleapis/python-genai#709

Interestingly, the Gen AI SDK surfaces more generation-time metadata than the Vertex AI SDK (though that’s unrelated to the serialization method itself).

---

Takeaways

• When moving from the Vertex AI SDK to the Gen AI SDK, remember that GenerateContentResponse is Pydantic-based.

  • Serialize with model_dump() or model_dump_json(), not to_dict().

• Some features are still exclusive to the Vertex AI SDK, so the Gen AI SDK isn’t yet a drop-in replacement for every use case.

• That said, Google’s docs now recommend the Gen AI SDK, and it’s seeing the most active development—so consolidation onto this SDK feels inevitable.

Looking forward to future updates!