Google

The GoogleModel is a model that uses the google-genai package under the hood to access Google’s Gemini models via both the Gemini API and Google Cloud (formerly known as Vertex AI).

Two providers wrap those endpoints:

• GoogleProvider — the Gemini API (Google AI Studio), surfaced under the 'google:' prefix.
• GoogleCloudProvider — Google Cloud (formerly known as Vertex AI), surfaced under the 'google-cloud:' prefix.

Both gateway/google:<model> and gateway/google-cloud:<model> use GoogleCloudProvider through Pydantic AI Gateway. Gemini API-only features, such as FileSearchTool, require the direct Gemini API through GoogleProvider.

Install

To use GoogleModel, you need to either install pydantic-ai, or install pydantic-ai-slim with the google optional group:

pip

Terminal

pip install "pydantic-ai-slim[google]"

uv

Terminal

uv add "pydantic-ai-slim[google]"

Configuration

GoogleModel lets you use Google’s Gemini models through their Gemini API (generativelanguage.googleapis.com) or Google Cloud (*-aiplatform.googleapis.com, formerly known as Vertex AI).

API Key (Gemini API)

To use Gemini via the Gemini API, go to aistudio.google.com and create an API key.

Once you have the API key, set it as an environment variable:

Terminal

export GOOGLE_API_KEY=your-api-key

You can then use GoogleModel by name:

from pydantic_ai import Agent

agent = Agent('google:gemini-3.8-flash')
...

Tip

For voice agents, use a Gemini Live model with a realtime session instead.

Or you can explicitly create the provider:

from pydantic_ai import Agent
from pydantic_ai.models.google import GoogleModel
from pydantic_ai.providers.google import GoogleProvider

provider = GoogleProvider(api_key='your-api-key')
model = GoogleModel('gemini-3.8-flash', provider=provider)
agent = Agent(model)
...

Google Cloud (Enterprise)

See Google Cloud for Gemini and Claude access, authentication, and Model Garden setup.

• Application Default Credentials
• Service Account
• API Key
• Customizing Location or Project

Service tier (service_tier, google_cloud_service_tier)

The unified service_tier field works on both Google subsystems, with google_cloud_service_tier available for finer Google Cloud routing control. The provider-specific field wins when both are set.

Gemini API — sent as the request’s service_tier field:

service_tier	Sent to Gemini API
'auto'	(omitted — server default)
'default'	'standard'
'flex'	'flex'
'priority'	'priority'

Google Cloud — sent as HTTP routing headers; 'flex' and 'priority' always pick the PT-with-spillover variant, so customers with Provisioned Throughput (PT) keep using their reserved capacity first:

service_tier	Google Cloud routing headers	Effective behavior
'auto' / 'default'	(none)	PT first, then standard on-demand spillover
'flex'	X-Vertex-AI-LLM-Shared-Request-Type: flex	PT first, then Flex PayGo spillover
'priority'	X-Vertex-AI-LLM-Shared-Request-Type: priority	PT first, then Priority PayGo spillover

To bypass PT entirely (or use it exclusively, or any of the other Google Cloud-specific routing combinations) set google_cloud_service_tier directly — the unified field is intentionally limited to the safe PT-with-spillover variants.

Google Cloud — full set of routing values

The full google_cloud_service_tier values map to these HTTP headers:

• 'pt_only': PT only (X-Vertex-AI-LLM-Request-Type: dedicated).
• 'pt_then_flex': PT when quota allows, then Flex PayGo spillover (X-Vertex-AI-LLM-Shared-Request-Type: flex).
• 'pt_then_priority': PT when quota allows, then Priority PayGo spillover (X-Vertex-AI-LLM-Shared-Request-Type: priority).
• 'on_demand': Standard on-demand only (X-Vertex-AI-LLM-Request-Type: shared).
• 'flex_only': Flex PayGo only (X-Vertex-AI-LLM-Request-Type: shared and X-Vertex-AI-LLM-Shared-Request-Type: flex).
• 'priority_only': Priority PayGo only (X-Vertex-AI-LLM-Request-Type: shared and X-Vertex-AI-LLM-Shared-Request-Type: priority).

Example

from pydantic_ai import Agent
from pydantic_ai.models.google import GoogleModel, GoogleModelSettings
from pydantic_ai.providers.google_cloud import GoogleCloudProvider

provider = GoogleCloudProvider(location='global')
model = GoogleModel('gemini-3.8-flash', provider=provider)
agent = Agent(model)

result = agent.run_sync(
    'Hello!',
    model_settings=GoogleModelSettings(google_cloud_service_tier='pt_then_flex'),
)

Swap 'pt_then_flex' for any GoogleCloudServiceTier value — e.g. 'pt_then_priority' for Priority PayGo spillover, or 'flex_only' / 'priority_only' to bypass PT entirely.

After the request, inspect ModelResponse provider_details.get('traffic_type') (e.g. ON_DEMAND_FLEX, ON_DEMAND_PRIORITY) to see which tier served it, when the API returns it.

Model Garden

See Google Cloud Model Garden.

Custom HTTP Client

Google providers use httpx2 by default. You can pass a custom httpx2.AsyncClient to control transport settings:

from httpx2 import AsyncClient

from pydantic_ai import Agent
from pydantic_ai.models.google import GoogleModel
from pydantic_ai.providers.google import GoogleProvider

custom_http_client = AsyncClient(timeout=30)
model = GoogleModel(
    'gemini-3.8-flash',
    provider=GoogleProvider(api_key='your-api-key', http_client=custom_http_client),
)
agent = Agent(model)
...

The client the Google providers create, including through the Gateway, has a 20-minute timeout, which long decode needs. For a client you pass, google-genai applies the client’s read timeout, or 600 seconds if it has none, to every phase of each request, so give it a read timeout of at least 20 minutes for long-decode models.

The Google providers also accept a legacy httpx.AsyncClient during Pydantic AI v2, but emit a deprecation warning. Use httpx2.AsyncClient for new code; legacy HTTPX client support will be removed in Pydantic AI v3.

HTTP Retries

Note

For most use cases, the model-agnostic transport retries approach is preferable, as it works the same way across all providers. The retry_options argument below is a Google-specific alternative that delegates retrying to the google-genai SDK’s own HTTP layer. See The layers in the retries guide for where these SDK-level retries sit, and Retry multiplication for how they compound with the agent’s own retry budgets.

By default, the google-genai SDK does not retry requests that fail with a transient HTTP error. You can enable retries by passing a HttpRetryOptions instance to the retry_options argument of GoogleProvider or GoogleCloudProvider:

from google.genai.types import HttpRetryOptions

from pydantic_ai import Agent
from pydantic_ai.models.google import GoogleModel
from pydantic_ai.providers.google import GoogleProvider

retry_options = HttpRetryOptions(
    attempts=4,
    initial_delay=1.0,
    max_delay=60.0,
    http_status_codes=[408, 429, 500, 502, 503, 504],
)
model = GoogleModel(
    'gemini-3.8-flash',
    provider=GoogleProvider(api_key='your-api-key', retry_options=retry_options),
)
agent = Agent(model)
...

This passes the options through to the SDK’s HttpOptions.retry_options. ModelSettings['max_retries'] overrides the number of attempts for a model, agent or run, keeping the delays and status codes set here, or the SDK’s defaults (a first delay of one second; status 408, 429, 500, 502, 503 and 504) when retry_options isn’t set. See the Vertex AI retry strategy documentation for guidance on choosing values.

Document, Image, Audio, and Video Input

GoogleModel supports multi-modal input, including documents, images, audio, and video.

YouTube video URLs can be passed directly to Google models:

youtube_input.py

from pydantic_ai import Agent, VideoUrl
from pydantic_ai.models.google import GoogleModel

agent = Agent(GoogleModel('gemini-3.8-flash'))
result = agent.run_sync(
    [
        'What is this video about?',
        VideoUrl(url='https://www.youtube.com/watch?v=dQw4w9WgXcQ'),
    ]
)
print(result.output)

Files can be uploaded via the Files API and passed as URLs:

file_upload.py

from pydantic_ai import Agent, DocumentUrl
from pydantic_ai.models.google import GoogleModel
from pydantic_ai.providers.google import GoogleProvider

provider = GoogleProvider()
file = provider.client.files.upload(file='pydantic-ai-logo.png')
assert file.uri is not None

agent = Agent(GoogleModel('gemini-3.8-flash', provider=provider))
result = agent.run_sync(
    [
        'What company is this logo from?',
        DocumentUrl(url=file.uri, media_type=file.mime_type),
    ]
)
print(result.output)

See the input documentation for more details and examples.

Agentic video processing

Agentic video processing lets Gemini inspect relevant video segments on demand. Without media_processing, Gemini uses its default static processing. Google recommends static processing for short clips, whole-video summaries, latency-sensitive queries, or frame-level precision across the clip; agentic processing for long videos or retrieval of specific moments or topics. Choose the processing mode for the workload.

For Gemini models that support agentic video processing, set media_processing on a video input:

agentic_video.py

from pydantic_ai import Agent, VideoUrl
from pydantic_ai.models.google import GoogleModel

agent = Agent(GoogleModel('gemini-3.7-flash'))
result = agent.run_sync(
    [
        'Summarize the important moments with timestamps.',
        VideoUrl(
            url='https://www.youtube.com/watch?v=dQw4w9WgXcQ',
            vendor_metadata={'media_processing': 'AGENTIC'},
        ),
    ]
)
print(result.output)

Video source and processing mode are independent. A YouTube VideoUrl uses Gemini’s direct YouTube input: Pydantic AI sends the URL without downloading or uploading the video, including when agentic processing is enabled. Google lists YouTube URLs among the supported agentic video sources.

On the Gemini Developer API, each processing step appears in the response as a NativeToolCallPart and NativeToolReturnPart named media_processing. Vertex AI returns signature-only parts for these steps, so they do not appear in the message history.

Follow-up video processing

Pydantic AI omits media-processing steps from later requests because Gemini currently rejects them. The ordinary final response and its thought signature remain in history. When that history includes the original video, follow-up requests process it again and can incur additional video-processing tokens.

Image generation

Use ImageGenerator with a google: image model for direct generation and reference-image editing through the Gemini API, or with a google-cloud: model to run the same models on Vertex AI:

google_image_generation.py

from pydantic_ai import ImageGenerator
from pydantic_ai.images.google import GoogleImageGenerationSettings

settings = GoogleImageGenerationSettings(
    google_image_config={'aspect_ratio': '1:1', 'image_size': '1K'}
)

gemini_api_generator = ImageGenerator('google:gemini-3.1-flash-lite-image', settings=settings)
vertex_generator = ImageGenerator('google-cloud:gemini-3.1-flash-image', settings=settings)

Construct GoogleImageGenerationModel with a GoogleCloudProvider to set the Vertex project and location explicitly.

The direct adapter accepts inline images and downloadable image URLs on both APIs, and forwards a reference the selected transport hosts itself as a fileData part instead of downloading it, exactly as GoogleModel does: on the Gemini Developer API that is a Files API URI, as an UploadedFile or an ImageUrl; on Vertex AI, where the Files API is not available, it is a Cloud Storage gs://bucket/path URI, as an UploadedFile whose file_id starts with gs:// or an ImageUrl whose URL does and that isn’t force_download. Vertex reads the object server-side, so a multi-megabyte reference never passes through your process, and neither transport accepts the other’s references: a Files API id on Vertex raises UserError. Which API a model talks to is read off the client, not the provider name, so a Vertex-backed client passed to GoogleProvider is treated as Vertex, and a Gemini Developer API client passed to GoogleCloudProvider keeps Files API support. See the image-generation guide for the common API and geometry behavior. The adapter requests an image-only response because ImageGenerator returns generated images rather than Gemini’s optional conversational text.

Every generated image carries an unconditional SynthID watermark. The Gemini 3 image models are thinking models: thinking is always on and billed, and its tokens are included in the result’s usage.

Model settings

You can customize model behavior using GoogleModelSettings:

from google.genai.types import HarmBlockThreshold, HarmCategory

from pydantic_ai import Agent
from pydantic_ai.models.google import GoogleModel, GoogleModelSettings

settings = GoogleModelSettings(
    temperature=0.2,
    max_tokens=1024,
    top_k=40,
    google_safety_settings=[
        {
            'category': HarmCategory.HARM_CATEGORY_HATE_SPEECH,
            'threshold': HarmBlockThreshold.BLOCK_LOW_AND_ABOVE,
        }
    ]
)
model = GoogleModel('gemini-3.8-flash')
agent = Agent(model, model_settings=settings)
...

Configure thinking

Use the provider-agnostic Thinking capability to enable thinking:

from pydantic_ai import Agent
from pydantic_ai.capabilities import Thinking

agent = Agent('google:gemini-3.8-flash', capabilities=[Thinking(effort='medium')])
...

For advanced usage, you can pass Google’s native thinking config through GoogleModelSettings.google_thinking_config:

from google.genai.types import ThinkingLevel

from pydantic_ai import Agent
from pydantic_ai.models.google import GoogleModel, GoogleModelSettings

model = GoogleModel('gemini-3.8-flash')
model_settings = GoogleModelSettings(google_thinking_config={'include_thoughts': True, 'thinking_level': ThinkingLevel.MEDIUM})
agent = Agent(model, model_settings=model_settings)
...

Pydantic AI resolves each model’s supported levels from Google’s documented thinking table and snaps a requested effort to the nearest supported level. gemini-3.1-flash-image is the exception: Google documents minimal, high for it, but only the Gemini API enforces them, so Pydantic AI applies those levels only when the client talks to the Gemini API, and Vertex AI keeps the full scale. For a model id the table doesn’t cover, declare its levels with GoogleModelProfile.google_thinking_levels (default: the full scale); unsupported efforts resolve to the nearest supported level.

gemini-4-argon supports the low, medium and high levels, so thinking=False and 'minimal' resolve to low. On Google Cloud it rejects a thinking_budget in google_thinking_config with a 400 error: use thinking_level instead.

thinking=False turns thinking off where the model allows it. The Pro models can’t turn it off, so they think as little as they allow instead: Gemini 3 Pro models use their lowest thinking level (low on gemini-3.1-pro-preview), and Gemini 2.5 Pro uses its minimum thinking_budget of 128 tokens. Neither asks for the thoughts to be returned. On Google Cloud, gemini-3.1-pro-preview accepts thinking_budget=0, which the Gemini API rejects for it, so there thinking=False sends thinking_budget=0 and the model doesn’t think.

See Thinking for the unified API and Gemini API docs for Google’s native thinking configuration.

Safety settings

You can customize the safety settings by setting the google_safety_settings field.

from google.genai.types import HarmBlockThreshold, HarmCategory

from pydantic_ai import Agent
from pydantic_ai.models.google import GoogleModel, GoogleModelSettings

model_settings = GoogleModelSettings(
    google_safety_settings=[
        {
            'category': HarmCategory.HARM_CATEGORY_HATE_SPEECH,
            'threshold': HarmBlockThreshold.BLOCK_LOW_AND_ABOVE,
        }
    ]
)
model = GoogleModel('gemini-3.8-flash')
agent = Agent(model, model_settings=model_settings)
...

See the Gemini API docs for more on safety settings.

Logprobs

You can return logprobs from the model in your response by setting google_logprobs and google_top_logprobs in the GoogleModelSettings.

This feature is only supported for non-streaming requests and Google Cloud.

from pydantic_ai import Agent
from pydantic_ai.models.google import GoogleModel, GoogleModelSettings
from pydantic_ai.providers.google_cloud import GoogleCloudProvider

model_settings = GoogleModelSettings(
    google_logprobs=True, google_top_logprobs=2,
)

model = GoogleModel(
    model_name='gemini-3.8-flash',
    provider=GoogleCloudProvider(location='europe-west1'),
)
agent = Agent(model, model_settings=model_settings)

result = agent.run_sync('Your prompt here')
# Access logprobs from provider_details
provider_details = result.response.provider_details or {}
logprobs = provider_details.get('logprobs')
avg_logprobs = provider_details.get('avg_logprobs')

See the Google Dev Blog for more information.

Model Armor (Google Cloud only)

Model Armor is a Google Cloud security service that screens prompts and responses for risks like prompt injection, jailbreaking, and sensitive data leakage.

You can configure it via google_model_armor_config in GoogleModelSettings:

from pydantic_ai import Agent
from pydantic_ai.models.google import GoogleModel, GoogleModelSettings
from pydantic_ai.providers.google_cloud import GoogleCloudProvider

model_settings = GoogleModelSettings(
    google_model_armor_config={
        'prompt_template_name': 'projects/my-project/locations/europe-west4/templates/prompt-template',
        'response_template_name': 'projects/my-project/locations/europe-west4/templates/response-template',
    }
)

model = GoogleModel(
    model_name='gemini-3.8-flash',
    provider=GoogleCloudProvider(location='europe-west4'),
)
agent = Agent(model, model_settings=model_settings)
...

Templates must be created in advance in the Google Cloud Console and must reside in the same region as the model endpoint. See the Model Armor Vertex AI integration docs for supported locations.

When a prompt or response is blocked, a ContentFilterError is raised.

Note that Model Armor screening — both prompt and response templates — only works with non-streaming requests (agent.run()). With streaming (agent.run_stream()), Google Cloud does not apply Model Armor: the prompt is not screened and the response text is returned unscreened. If you require streaming and need Model Armor protection, pre-screen prompts using the google-cloud-modelarmor SDK before calling the agent.

Context caching (google_cached_content)

When you’ve created a Gemini cached content resource, pass its resource name through google_cached_content to reuse it across requests. (Gemini also caches prompts implicitly, so the unified ModelSettings.cache setting, which configures request-side caching on other providers, adds nothing to a Google request.)

from pydantic_ai import Agent
from pydantic_ai.models.google import GoogleModel, GoogleModelSettings

model_settings = GoogleModelSettings(
    google_cached_content='projects/p/locations/global/cachedContents/your-cache-id',
)

agent = Agent(GoogleModel('gemini-3.8-flash'), model_settings=model_settings)
...

Cached fields are owned by the cache resource

The cache resource owns system_instruction, tools, and tool_config — Pydantic AI strips them from outgoing requests when google_cached_content is set, so agent instructions and registered tools are ignored on cached requests. A UserWarning is emitted whenever stripping drops a field, so the mismatch is discoverable.

Create a cached content resource

Pydantic AI doesn’t wrap the cache-management API — create the resource with the underlying google-genai SDK, then pass its name through google_cached_content:

from google.genai.types import Content, CreateCachedContentConfig, Part

from pydantic_ai.providers.google import GoogleProvider

provider = GoogleProvider(api_key='your-api-key')

cache = provider.client.caches.create(
    model='gemini-3.8-flash',
    config=CreateCachedContentConfig(
        system_instruction='You are a geography expert. Be concise.',
        contents=[Content(role='user', parts=[Part(text='...long context to cache...')])],
        ttl='3600s',
    ),
)
print(cache.name)
#> cachedContents/abc123...

Caches have a minimum size (≈2048 tokens for Gemini 2.5 series models, ≈4096 for Gemini 3 series models) and a TTL — see the Gemini caching docs for the current thresholds, pricing, and list / update / delete operations.

Long decode

gemini-4-argon caps the output of each request; Google’s long decode docs give the caps and defaults. When max_tokens is above the cap, a request that reaches it returns a continuation token, and the google-genai SDK continues the response from it in another request, until the response is done or max_tokens is spent:

from pydantic_ai import Agent

agent = Agent('google:gemini-4-argon', model_settings={'max_tokens': 100_000})
result = agent.run_sync('Write the complete specification.')
print(result.output)

The SDK merges the requests into one ModelResponse, so to Pydantic AI a long decode is one model request:

• Its usage is the sum over the requests, which is what Google bills. Each request’s input counts the output so far again.
• Usage limits count it as one request, and a run that isn’t streamed checks its token limits once every request is done: max_tokens is what bounds a long decode. per_request_input_tokens_limit is checked against the input tokens of the first request.
• Timeouts apply to each request. The Google providers create their client with the 20-minute read timeout Google recommends.
• If a later request fails, the model request fails. A run that isn’t streamed loses the output of the earlier requests, already billed. A streamed run keeps what it streamed as a response with state='interrupted' that carries the finished requests’ usage, which capture_run_messages() returns.
• The SDK resumes a dropped stream from its last continuation token. If output was streamed after that token, the model request fails as above instead, since the resumed request would repeat that output.
• On Temporal, the whole long decode runs in one model request activity, so its start_to_close_timeout must cover every request. A retried activity starts the response over. The merged response is the activity’s one result, so a very long output can exceed Temporal’s payload limit.

Without max_tokens, Google’s default output limit applies, and a response that reaches it ends with finish_reason='length'.

On Google Cloud, gemini-4-argon is served on the global location and the us and eu multi-regions, not in single regions such as the default us-central1: pass one of those as the provider’s location.

Streaming cancellation

Transport cancellation

cancel() safely interrupts an active local stream pull, including one running in another task. The google-genai SDK exposes no documented per-stream transport handle, so closing the returned iterator does not guarantee immediate HTTP teardown or indicate when remote generation stops and billing ends. See googleapis/python-genai#2425.