Gemini API Complete Guide: Pricing, Model Selection, Multimodal, Function Calling

The Gemini API is Google's latest multimodal LLM family, with the Gemini 2.0 generation rolling out from 2024 onward. Three tiers (Flash / Pro / Ultra) deliver 75% cost reduction via Context Cache, up to 2 million tokens of context, and built-in Tool Use, placing it alongside OpenAI and Anthropic as a top-tier LLM API.

Model Family (2026)

Google AI Studio vs Vertex AI

Key Features

• Multimodal: text + image + audio + video + PDF
• Long Context: 1M / 2M tokens
• Function Calling: Tool Use, parallel execution, Strict mode
• Code Execution: sandboxed Python execution (Pro)
• Context Caching: 75% input token discount
• Grounding with Google Search: cite up-to-date information
• JSON Mode: enforce structured output
• Streaming: incremental responses via SSE
• Batch API: async with 50% discount

Python Sample (Vertex AI SDK)

from vertexai.generative_models import GenerativeModel, Part
import vertexai

vertexai.init(project="my-project", location="asia-northeast1")
model = GenerativeModel("gemini-2.0-flash-001")

# Text
response = model.generate_content("What is the capital of Japan?")
print(response.text)

# Multimodal (image + text)
image = Part.from_uri("gs://my-bucket/photo.jpg", mime_type="image/jpeg")
response = model.generate_content([image, "Describe this image"])

# Function Calling
from vertexai.generative_models import Tool, FunctionDeclaration

weather_func = FunctionDeclaration(
    name="get_weather",
    description="Get current weather",
    parameters={
        "type": "object",
        "properties": {"city": {"type": "string"}},
        "required": ["city"],
    },
)
tools = [Tool(function_declarations=[weather_func])]
response = model.generate_content(
    "What is the weather in Tokyo?", tools=tools
)

Context Caching Pattern

from vertexai.preview import caching
from datetime import datetime, timedelta

# Cache the system prompt plus a long document
cached_content = caching.CachedContent.create(
    model_name="gemini-2.0-pro-001",
    system_instruction="You are an internal FAQ bot",
    contents=[long_document],
    ttl=timedelta(hours=1),
)

# Using the cache (75% discount on input)
model = GenerativeModel.from_cached_content(cached_content)
response = model.generate_content("How do I request time off?")

Comparison with Other LLM APIs

Typical Use Cases

• RAG (internal document search + answering)
• Video content analysis and summarization
• Multimodal OCR + structured extraction
• Code generation + refactoring
• Agents (external API integration via Tool Use)
• Multilingual translation (100+ languages)

* Google and Gemini are trademarks of Google LLC. For the latest information, see the Vertex AI Gen AI official docs.