Skip to main content

​
Gemini Models Overview

Gemini is a family of multimodal generative AI models developed by Google DeepMind, designed for state-of-the-art performance across text, code, images, audio, and video understanding.

​
Model Family

The Gemini family includes several model variants optimized for different use cases:

​
Gemini 3.1 Pro

Gemini 3.1 Pro is Google’s latest flagship model with enhanced capabilities:
  • Enhanced stability & grounding: Improved factuality and reduced repetitive response patterns
  • Advanced coding & agentic capabilities: Significant improvements in software engineering and agentic performance
  • Efficiency & reasoning modes: Improved token efficiency with Medium thinking level to balance depth with latency
  • Core quality improvements: Advanced reasoning, instruction following, and creative writing

​
Gemini 3 Flash

Optimized for speed and efficiency while maintaining high-quality output:
  • Fast response times for real-time applications
  • Cost-effective for high-volume workloads
  • Supports multimodal inputs (text, images, video, audio)
  • Excellent for chat, code generation, and content analysis

​
Gemini 2.5 Models

Previous generation models with strong performance:
  • Gemini 2.5 Pro: Balanced performance for complex reasoning
  • Gemini 2.5 Flash: Speed-optimized variant
  • Gemini 2.5 Flash Lite: Ultra-lightweight for edge deployments

​
Key Capabilities

Multimodal Understanding

Process and understand text, images, video, audio, and PDFs in a single unified model

Advanced Reasoning

Dynamic thinking with configurable reasoning depth for complex problem-solving

Code Generation

Generate, execute, and debug Python code with built-in code execution

Function Calling

Integrate with external APIs and tools through structured function declarations

Grounding

Connect to real-time data via Google Search or custom data sources

Long Context

Process up to millions of tokens with context caching for cost optimization

​
Available Models

Model IDContext WindowStrengthsBest For
gemini-3.1-pro-preview2M tokensAdvanced reasoning, codingComplex tasks, agents
gemini-3-flash-preview1M tokensSpeed, efficiencyReal-time apps, high volume
gemini-2.5-pro2M tokensBalanced performanceGeneral purpose
gemini-2.5-flash1M tokensFast responsesProduction workloads
gemini-2.5-flash-lite1M tokensLightweightEdge deployment

​
Getting Started

1

Install the SDK

Install the Google Gen AI SDK for Python:
pip install --upgrade google-genai
2

Set up authentication

Authenticate your environment and enable the Vertex AI API:
from google.colab import auth
auth.authenticate_user()
3

Initialize the client

Create a client with your project details:
from google import genai

client = genai.Client(
    vertexai=True,
    project="your-project-id",
    location="global"
)
4

Generate content

Send your first request:
response = client.models.generate_content(
    model="gemini-3.1-pro-preview",
    contents="Explain how AI works in simple terms."
)
print(response.text)

​
Model Configuration

Control model behavior with generation parameters: 
from google.genai.types import GenerateContentConfig, ThinkingConfig, ThinkingLevel

response = client.models.generate_content(
    model="gemini-3.1-pro-preview",
    contents="How does AI work?",
    config=GenerateContentConfig(
        temperature=1.0,          # Creativity (0.0-2.0)
        top_p=0.95,              # Nucleus sampling
        max_output_tokens=8000,  # Response length limit
        thinking_config=ThinkingConfig(
            thinking_level=ThinkingLevel.LOW  # Reasoning depth
        )
    )
)
For Gemini 3 models, we recommend keeping temperature at the default value of 1.0 as the reasoning capabilities are optimized for this setting.

​
Thinking Levels

Gemini 3.1 Pro introduces granular control over reasoning depth:
  • Low (1-1K tokens): Minimizes latency for simple tasks
  • Medium (1K-16K tokens): Balances reasoning and latency
  • High (16K-32K tokens): Maximizes reasoning depth (default)
from google.genai.types import ThinkingConfig, ThinkingLevel

# For faster responses on simple queries
config = GenerateContentConfig(
    thinking_config=ThinkingConfig(
        thinking_level=ThinkingLevel.LOW
    )
)

# For complex reasoning tasks
config = GenerateContentConfig(
    thinking_config=ThinkingConfig(
        thinking_level=ThinkingLevel.HIGH
    )
)

​
Media Resolution Control

Gemini 3 allows fine-grained control over image and video processing: 
from google.genai.types import Part, PartMediaResolution, PartMediaResolutionLevel

response = client.models.generate_content(
    model="gemini-3.1-pro-preview",
    contents=[
        Part(
            file_data=types.FileData(
                file_uri="gs://path/to/image.jpg",
                mime_type="image/jpeg"
            ),
            media_resolution=PartMediaResolution(
                level=PartMediaResolutionLevel.MEDIA_RESOLUTION_HIGH
            )
        ),
        "Analyze this image in detail."
    ]
)

​
Safety Settings

Configure content filtering for responsible AI: 
from google.genai.types import SafetySetting, HarmCategory, HarmBlockThreshold

safety_settings = [
    SafetySetting(
        category=HarmCategory.HARM_CATEGORY_HARASSMENT,
        threshold=HarmBlockThreshold.BLOCK_LOW_AND_ABOVE
    ),
    SafetySetting(
        category=HarmCategory.HARM_CATEGORY_HATE_SPEECH,
        threshold=HarmBlockThreshold.BLOCK_LOW_AND_ABOVE
    )
]

response = client.models.generate_content(
    model="gemini-3.1-pro-preview",
    contents="Your prompt here",
    config=GenerateContentConfig(safety_settings=safety_settings)
)

​
Next Steps

Getting Started

Learn basic text generation and model configuration

Multimodal Inputs

Process images, video, audio, and documents

Function Calling

Connect Gemini to external tools and APIs

Grounding

Ground responses in Google Search or custom data

​
Resources

Build docs developers (and LLMs) love

Get started for freeTalk to us