Google Gemini Cheat Sheet

Updated 2026-05-28

🧠Study flashcards on this topic113 cards · spaced repetition→

Google Gemini is a family of multimodal large language models by Google DeepMind, capable of understanding and generating text, images, audio, video, and code. As of May 2026, Gemini 3.5 Flash (released at Google I/O 2026) leads the stable lineup as the strongest agentic and coding model — outperforming Gemini 3.1 Pro on key benchmarks at 4× the speed of other frontier models — while Gemini 3.1 Pro Preview (2M token context) remains the flagship for deep reasoning. The API is accessible via two paths: the Gemini Developer API (API keys, Google AI Studio) and Vertex AI (enterprise GCP deployment with IAM). The official Python SDK is google-genai (pip install google-genai), offering both the classic generate_content and the new Interactions API (client.interactions.create()) designed for agentic workflows.

What This Cheat Sheet Covers

This topic spans 17 focused tables and 134 indexed concepts, 113 flashcards. Below is a complete table-by-table outline of this topic, spanning foundational concepts through advanced details.

Table 1: Model VariantsTable 2: Multimodal CapabilitiesTable 3: API SDKs and IntegrationTable 4: Prompting and ConfigurationTable 5: Function Calling and Tool UseTable 6: Context and Memory ManagementTable 7: Safety and Content FilteringTable 8: Advanced FeaturesTable 9: API Access and PricingTable 10: Rate Limits and QuotasTable 11: Authentication and SecurityTable 12: Prompt Engineering TechniquesTable 13: Error Handling and TroubleshootingTable 14: Use Cases and ApplicationsTable 15: Comparison with Other ModelsTable 16: Developer Tools and ResourcesTable 17: Generative Media Models

Quick IndexSubscribe to unlock

A jump-to index of every table row in this cheat sheet.

Mind MapSubscribe to unlock

An interactive map of every table and concept in this topic.

Table 1: Model Variants

Each Gemini generation brings a different trade-off of intelligence, speed, and cost; choosing the right model — and knowing which are stable versus preview versus shutting down — is the single most important configuration decision in any Gemini project.

Model	Example	Description
Gemini 3.5 Flash	`model='gemini-3.5-flash'`	• Current stable flagship — released May 19, 2026 (Google I/O 2026) • outperforms Gemini 3.1 Pro on coding and agentic benchmarks; 4× faster than other frontier models • 1M token context, 64K max output; thinking levels `minimal`/`low`/`medium` (default)/`high` • free tier available; Computer Use not yet supported.
Gemini 3.1 Pro Preview	`model='gemini-3.1-pro-preview'`	• Flagship for complex reasoning, multimodal, and vibe-coding tasks • 2M token context window — largest of any production model • `thinking_level` enum: `LOW`, `MEDIUM`, `HIGH`; no free tier • separate `gemini-3.1-pro-preview-customtools` endpoint for prioritizing custom tools.
Gemini 3.1 Flash-Lite	`model='gemini-3.1-flash-lite'`	• Most cost-efficient stable model (GA May 7, 2026; replaces Preview) • optimized for high-volume agentic tasks, translation, simple data processing • free tier available; shutdown May 7, 2027.
Gemini 3 Flash Preview	`model='gemini-3-flash-preview'`	• Earlier-generation preview — recommended migration to `gemini-3.5-flash` • Agentic Vision: zooms, inspects, and manipulates images via code execution • `thinking_level` for latency control; free tier available; no shutdown date announced.
Gemini 2.5 Flash	`model='gemini-2.5-flash'`	• Widely-used stable model — best price-performance for low-latency, high-volume reasoning • 1M token context, 65K max output; `thinking_budget` or `thinking_level` • free tier available; shuts down October 16, 2026; migrate to `gemini-3.5-flash`.
Gemini 2.5 Pro	`model='gemini-2.5-pro'`	• Stable Pro for complex reasoning and coding tasks • 1M token context, 65K max output; adaptive thinking, implicit context caching • shuts down October 16, 2026; migrate to `gemini-3.1-pro-preview`.
Gemini 2.5 Flash-Lite	`model='gemini-2.5-flash-lite'`	• Budget stable model for high-throughput, latency-sensitive workloads • 1M token context; free tier available • shuts down October 16, 2026; migrate to `gemini-3.1-flash-lite`.
Gemini Nano 4	AICore system service on Android Not accessible via Gemini API	• On-device edge AI for Android (developer preview) • two variants: Nano 4 Fast (E2B) and Nano 4 Full (E4B) • 140+ language support; 4× faster and 60% less battery than predecessor; runs offline.

Table 2: Multimodal Capabilities

Gemini's native multimodal architecture means all modalities — text, images, audio, video, documents — share the same token budget and can be freely combined in a single request; this is fundamentally different from models that bolt on vision or audio as separate modules.

Capability	Example	Description
Text generation	`response = client.models.generate_content(` `model='gemini-3.5-flash',` `contents='Explain quantum computing')`	• Natural language understanding and generation for conversations, summaries, and translations • supports multi-turn chat with persistent context • system instructions for behavior steering.

The rest of this section is part of the full cheat sheet

Table 3: API SDKs and Integration

The google-genai SDK is the single entry point for both the classic generate_content API and the new Interactions API; the same package targets both the Gemini Developer API and Vertex AI with only the Client initialization differing.

SDK	Example	Description
Python SDK (google-genai)	`pip install --upgrade google-genai` `from google import genai` `from google.genai import types` `client = genai.Client(api_key='YOUR_KEY')` `r = client.models.generate_content(` `model='gemini-3.5-flash', contents='Hi')`	• Official current SDK for Python 3.10+ • unified `Client` object for all Gemini API calls • same code targets both Gemini Developer API and Vertex AI (swap `Client` init) • v2.0.0+ required for full Interactions API support.

The rest of this section is part of the full cheat sheet

Table 4: Prompting and Configuration

Gemini 3.x models are optimized for their default settings; for Gemini 3.x, thinking_level replaces numeric thinking_budget, and temperature/top_p/top_k are no longer recommended — these settings can degrade reasoning quality on 3.x models.

Parameter	Example	Description
System instructions	`config=types.GenerateContentConfig(` `system_instruction='You are a Python expert')`	• Steers model behavior across the entire conversation • sets role, tone, and constraints; applies to all turns • not supported by image generation models.

The rest of this section is part of the full cheat sheet

Table 1: Model Variants

Model	Example	Description
Gemini 3.5 Flash	`model='gemini-3.5-flash'`	• Current stable flagship — released May 19, 2026 (Google I/O 2026) • outperforms Gemini 3.1 Pro on coding and agentic benchmarks; 4× faster than other frontier models • 1M token context, 64K max output; thinking levels `minimal`/`low`/`medium` (default)/`high` • free tier available; Computer Use not yet supported.
Gemini 3.1 Pro Preview	`model='gemini-3.1-pro-preview'`	• Flagship for complex reasoning, multimodal, and vibe-coding tasks • 2M token context window — largest of any production model • `thinking_level` enum: `LOW`, `MEDIUM`, `HIGH`; no free tier • separate `gemini-3.1-pro-preview-customtools` endpoint for prioritizing custom tools.
Gemini 3.1 Flash-Lite	`model='gemini-3.1-flash-lite'`	• Most cost-efficient stable model (GA May 7, 2026; replaces Preview) • optimized for high-volume agentic tasks, translation, simple data processing • free tier available; shutdown May 7, 2027.
Gemini 3 Flash Preview	`model='gemini-3-flash-preview'`	• Earlier-generation preview — recommended migration to `gemini-3.5-flash` • Agentic Vision: zooms, inspects, and manipulates images via code execution • `thinking_level` for latency control; free tier available; no shutdown date announced.
Gemini 2.5 Flash	`model='gemini-2.5-flash'`	• Widely-used stable model — best price-performance for low-latency, high-volume reasoning • 1M token context, 65K max output; `thinking_budget` or `thinking_level` • free tier available; shuts down October 16, 2026; migrate to `gemini-3.5-flash`.
Gemini 2.5 Pro	`model='gemini-2.5-pro'`	• Stable Pro for complex reasoning and coding tasks • 1M token context, 65K max output; adaptive thinking, implicit context caching • shuts down October 16, 2026; migrate to `gemini-3.1-pro-preview`.
Gemini 2.5 Flash-Lite	`model='gemini-2.5-flash-lite'`	• Budget stable model for high-throughput, latency-sensitive workloads • 1M token context; free tier available • shuts down October 16, 2026; migrate to `gemini-3.1-flash-lite`.
Gemini Nano 4	AICore system service on Android Not accessible via Gemini API	• On-device edge AI for Android (developer preview) • two variants: Nano 4 Fast (E2B) and Nano 4 Full (E4B) • 140+ language support; 4× faster and 60% less battery than predecessor; runs offline.

Table 2: Multimodal Capabilities

Capability	Example	Description
Text generation	`response = client.models.generate_content(` `model='gemini-3.5-flash',` `contents='Explain quantum computing')`	• Natural language understanding and generation for conversations, summaries, and translations • supports multi-turn chat with persistent context • system instructions for behavior steering.

Table 3: API SDKs and Integration

SDK	Example	Description
Python SDK (google-genai)	`pip install --upgrade google-genai` `from google import genai` `from google.genai import types` `client = genai.Client(api_key='YOUR_KEY')` `r = client.models.generate_content(` `model='gemini-3.5-flash', contents='Hi')`	• Official current SDK for Python 3.10+ • unified `Client` object for all Gemini API calls • same code targets both Gemini Developer API and Vertex AI (swap `Client` init) • v2.0.0+ required for full Interactions API support.

Table 4: Prompting and Configuration

Parameter	Example	Description
System instructions	`config=types.GenerateContentConfig(` `system_instruction='You are a Python expert')`	• Steers model behavior across the entire conversation • sets role, tone, and constraints; applies to all turns • not supported by image generation models.