from google import genai  # CRITICAL: NOT google.generativeai

client = genai.Client()  # Uses GEMINI_API_KEY env var automatically

response = client.models.generate_content(
    model="gemini-3-flash-preview",
    contents="Hello world"
)
print(response.text)

import { GoogleGenAI } from "@google/genai";
const ai = new GoogleGenAI({});  // Uses GEMINI_API_KEY env var
const response = await ai.models.generateContent({
    model: "gemini-3-flash-preview",
    contents: "Hello world"
});
console.log(response.text);

pip install google-genai          # Python
npm install @google/genai          # JavaScript
go get google.golang.org/genai     # Go

**OpenAI compatibility** (3 line change):
```python
from openai import OpenAI
client = OpenAI(
    api_key="GEMINI_API_KEY",
    base_url="https://generativelanguage.googleapis.com/v1beta/openai/"
)
response = client.chat.completions.create(model="gemini-3-flash-preview", messages=[...])

from google.genai import types

response = client.models.generate_content(
    model="gemini-3-flash-preview",
    contents="User message",
    config=types.GenerateContentConfig(
        system_instruction="You are a helpful assistant.",
        temperature=1.0,
        max_output_tokens=1024,
    )
)

chat = client.chats.create(model="gemini-3-flash-preview")
response = chat.send_message("Hello")
print(response.text)
response2 = chat.send_message("Tell me more")
print(response2.text)

for chunk in client.models.generate_content_stream(
    model="gemini-3-flash-preview",
    contents="Write a long story"
):
    print(chunk.text, end="")

undefined

1 token ≈ 4 characters; 100 tokens ≈ 60-80 English words.

---

config=types.GenerateContentConfig(
    thinking_config=types.ThinkingConfig(thinking_level="low")  # "low", "medium", "high"
)

config=types.GenerateContentConfig(
    thinking_config=types.ThinkingConfig(thinking_budget=1024)  # token budget; 0=disabled
)

with open('image.jpg', 'rb') as f:
    image_bytes = f.read()

response = client.models.generate_content(
    model="gemini-3-flash-preview",
    contents=[
        types.Part.from_bytes(data=image_bytes, mime_type='image/jpeg'),
        "Caption this image."
    ]
)

import requests
image_bytes = requests.get("https://example.com/image.jpg").content
image = types.Part.from_bytes(data=image_bytes, mime_type="image/jpeg")

import pathlib
filepath = pathlib.Path('file.pdf')
response = client.models.generate_content(
    model="gemini-3-flash-preview",
    contents=[
        types.Part.from_bytes(data=filepath.read_bytes(), mime_type='application/pdf'),
        "Summarize this document"
    ]
)

myfile = client.files.upload(file="sample.mp3")
response = client.models.generate_content(
    model="gemini-3-flash-preview",
    contents=["Describe this audio", myfile]
)

myfile = client.files.upload(file="video.mp4")
response = client.models.generate_content(
    model="gemini-3-flash-preview",
    contents=[myfile, "Summarize this video"]
)

undefined

---

undefined

**Files API limits:** 2GB per file, 20GB per project, 48-hour TTL.

---

from pydantic import BaseModel

class Recipe(BaseModel):
    name: str
    ingredients: list[str]
    steps: list[str]

response = client.models.generate_content(
    model="gemini-3-flash-preview",
    contents="Give me a chocolate cake recipe",
    config=types.GenerateContentConfig(
        response_mime_type="application/json",
        response_schema=Recipe,
    )
)
import json
recipe = json.loads(response.text)

grounding_tool = types.Tool(google_search=types.GoogleSearch())
config = types.GenerateContentConfig(tools=[grounding_tool])
response = client.models.generate_content(
    model="gemini-3-flash-preview",
    contents="Who won Euro 2024?",
    config=config
)

Response includes `groundingMetadata` with `webSearchQueries`, `groundingChunks`, `groundingSupports`.

config = types.GenerateContentConfig(tools=[{"url_context": {}}])

undefined

undefined

undefined

config = types.GenerateContentConfig(
    tools=[types.Tool(code_execution=types.CodeExecution())]
)

def get_weather(location: str) -> dict:
    return {"temp": 72, "condition": "sunny"}

response = client.models.generate_content(
    model="gemini-3-flash-preview",
    contents="What's the weather in NYC?",
    config=types.GenerateContentConfig(
        tools=[get_weather],
        tool_config=types.ToolConfig(
            function_calling_config=types.FunctionCallingConfig(mode="AUTO")
        )
    )
)

**⚠️ Gemini 3 Thought Signatures in Function Calling:**  
When Gemini 3 returns function calls, each step includes a `thoughtSignature`. You MUST echo it back exactly — omitting it causes a 400 error. **The SDK handles this automatically if you use the chat API or append the full response object to history.**

Manual handling pattern:
```python

---

result = client.models.embed_content(
    model="gemini-embedding-001",
    contents="What is the meaning of life?"
)
print(result.embeddings)

Model: `gemini-embedding-001` (GA until July 14, 2026)  
Use case: semantic search, RAG, classification, clustering.

---

undefined

---

from google.genai import types

Implicit caching: 2048+ token prefix is automatically cached at 75% discount.  
Explicit caching: manual TTL control.

---

undefined

---

import time

interaction = client.interactions.create(
    input="Research the history of quantum computing",
    agent='deep-research-pro-preview-12-2025',
    background=True  # REQUIRED for long tasks
)

while True:
    interaction = client.interactions.get(interaction.id)
    if interaction.status == "completed":
        print(interaction.outputs[-1].text)
        break
    elif interaction.status == "failed":
        break
    time.sleep(10)

interaction = client.interactions.create(
    input="Compare our Q4 report to public benchmarks",
    agent="deep-research-pro-preview-12-2025",
    background=True,
    tools=[{"type": "file_search", "file_search_store_names": ["fileSearchStores/my-store"]}]
)

import asyncio
from google import genai

client = genai.Client()
model = "gemini-2.5-flash-native-audio-preview-12-2025"

async def main():
    async with client.aio.live.connect(
        model=model,
        config={"response_modalities": ["AUDIO"]}
    ) as session:
        await session.send_client_content(
            turns="Hello, how are you?",
            turn_complete=True
        )
        async for response in session.receive():
            if response.data:
                # raw PCM audio bytes (24kHz, 16-bit, little-endian)
                pass
            if response.server_content and response.server_content.turn_complete:
                break

asyncio.run(main())

config=types.LiveConnectConfig(
    session_resumption=types.SessionResumptionConfig(handle=previous_handle)
)

**Context window compression (for long sessions):**
```python
config=types.LiveConnectConfig(
    context_window_compression=types.ContextWindowCompressionConfig(
        sliding_window=types.SlidingWindow()
    )
)

undefined

---

import datetime
now = datetime.datetime.now(tz=datetime.timezone.utc)

client = genai.Client(http_options={'api_version': 'v1alpha'})
token = client.auth_tokens.create(config={
    'uses': 1,
    'expire_time': now + datetime.timedelta(minutes=30),
    'new_session_expire_time': now + datetime.timedelta(minutes=1),
    'http_options': {'api_version': 'v1alpha'},
})

Can lock token to specific config:
```python
'live_connect_constraints': {
    'model': 'gemini-2.5-flash-native-audio-preview-12-2025',
    'config': {'response_modalities': ['AUDIO']}
}

from PIL import Image

response = client.models.generate_content(
    model="gemini-3.1-flash-image-preview",
    contents="Create a picture of a tropical beach at sunset"
)

for part in response.parts:
    if part.text:
        print(part.text)
    elif part.inline_data:
        image = part.as_image()
        image.save("output.png")

operation = client.models.generate_videos(
    model="veo-3.1-generate-preview",
    prompt="A serene mountain lake at dawn"
)

while not operation.done:
    time.sleep(10)
    operation = client.operations.get(operation)

video = operation.response.generated_videos[0]
client.files.download(file=video.video)
video.video.save("output.mp4")

response = client.models.generate_images(
    model='imagen-4.0-generate-001',
    prompt='Robot holding a red skateboard',
    config=types.GenerateImagesConfig(number_of_images=4)
)
for gen_image in response.generated_images:
    gen_image.image.show()

import wave

response = client.models.generate_content(
    model="gemini-2.5-flash-preview-tts",
    contents="Say cheerfully: Have a wonderful day!",
    config=types.GenerateContentConfig(
        response_modalities=["AUDIO"],
        speech_config=types.SpeechConfig(
            voice_config=types.VoiceConfig(
                prebuilt_voice_config=types.PrebuiltVoiceConfig(voice_name='Kore')
            )
        )
    )
)

data = response.candidates[0].content.parts[0].inline_data.data
with wave.open("out.wav", "wb") as wf:
    wf.setnchannels(1)
    wf.setsampwidth(2)
    wf.setframerate(24000)
    wf.writeframes(data)

config=types.GenerateContentConfig(
    response_modalities=["AUDIO"],
    speech_config=types.SpeechConfig(
        multi_speaker_voice_config=types.MultiSpeakerVoiceConfig(
            speaker_voice_configs=[
                types.SpeakerVoiceConfig(
                    speaker='Joe',
                    voice_config=types.VoiceConfig(
                        prebuilt_voice_config=types.PrebuiltVoiceConfig(voice_name='Kore')
                    )
                ),
                types.SpeakerVoiceConfig(
                    speaker='Jane',
                    voice_config=types.VoiceConfig(
                        prebuilt_voice_config=types.PrebuiltVoiceConfig(voice_name='Aoede')
                    )
                ),
            ]
        )
    )
)

client = genai.Client(http_options={'api_version': 'v1alpha'})

async with client.aio.live.music.connect(model='models/lyria-realtime-exp') as session:
    await session.set_weighted_prompts(prompts=[
        types.WeightedPrompt(text='minimal techno', weight=1.0)
    ])
    await session.set_music_generation_config(
        config=types.LiveMusicGenerationConfig(bpm=90, temperature=1.0)
    )
    await session.play()
    
    # Receive audio chunks
    async for message in session.receive():
        audio_data = message.server_content.audio_chunks[0].data
        # process PCM audio...

config = genai.types.GenerateContentConfig(
    tools=[types.Tool(
        computer_use=types.ComputerUse(
            environment=types.Environment.ENVIRONMENT_BROWSER,
            excluded_predefined_functions=["drag_and_drop"]  # optional
        )
    )]
)

response = client.models.generate_content(
    model='gemini-2.5-computer-use-preview-10-2025',
    contents=[{"role": "user", "parts": [{"text": "Search Amazon for wireless headphones"}]}],
    config=config
)

**Safety:** Check `safety_decision` in response — `require_confirmation` means pause before executing.

---

response = client.models.generate_content(
    model="gemini-3-flash-preview",
    contents="Your prompt",
    config=types.GenerateContentConfig(
        safety_settings=[
            types.SafetySetting(
                category=types.HarmCategory.HARM_CATEGORY_HATE_SPEECH,
                threshold=types.HarmBlockThreshold.BLOCK_LOW_AND_ABOVE
            )
        ]
    )
)

config = types.GenerateContentConfig(
    media_resolution=types.MediaResolution.MEDIA_RESOLUTION_HIGH  # LOW, MEDIUM, HIGH
)

client = genai.Client(http_options={'api_version': 'v1alpha'})
image_part = types.Part.from_bytes(
    data=image_bytes, mime_type='image/jpeg',
    media_resolution=types.MediaResolution.MEDIA_RESOLUTION_HIGH
)

client = genai.Client(http_options={'api_version': 'v1'})  # force stable

export GEMINI_API_KEY=your_key_here

from crewai import LLM
gemini_llm = LLM(model='gemini/gemini-3-flash-preview', api_key=api_key, temperature=1.0)

from langchain_google_genai import ChatGoogleGenerativeAI
llm = ChatGoogleGenerativeAI(model="gemini-3-flash-preview")

from llama_index.llms.google_genai import GoogleGenAI
llm = GoogleGenAI(model="gemini-3-flash-preview")

npm install ai @ai-sdk/google