Google Gemini Guide

What is this about?

"Gemini" is not one thing. It is a consumer app, a free developer playground, a programmatic API, an enterprise platform on Google Cloud, a layer inside Google Workspace, an IDE coding assistant, and a family of open-weight models (Gemma). This guide maps the surfaces, shows the shortest path to a first API call, and helps you pick the right entry point.

Source scope as of June 23, 2026

Based on official Google sources (ai.google.dev, aistudio.google.com, cloud.google.com, developers.google.com, workspace.google.com). This is a fast-moving stack — two renames matter and are reflected below: Vertex AI is now the "Gemini Enterprise Agent Platform", and Gemini Code Assist for individuals is moving to "Google Antigravity" (from June 18, 2026). Exact per-model token limits for the Gemini 3.x line are not published on the models overview page and are marked as unverified where they appear. Re-check anything dated within the last ~30 days against the live page before relying on it.

1. The mental model

Surface	What it is for	Primary user
Gemini app	Consumer AI assistant — chat, Deep Research, Gems, Canvas, agentic tasks	Consumers, knowledge workers
Google AI Studio	Free browser playground to try models, prototype prompts, and get an API key	Developers (fastest start)
Gemini API (Gemini Developer API)	Programmatic model access via an API key	Developers building apps
Gemini Enterprise Agent Platform (formerly Vertex AI)	Production platform on GCP — build, scale, and govern agents and models; Model Garden, MLOps, data-residency controls	Enterprises, production teams
Gemini in Google Workspace	Gemini built into Gmail, Docs, Sheets, Slides, Meet	Workspace business users
Gemini Code Assist / Google Antigravity	AI coding in VS Code, JetBrains, Cloud Shell (chat + agent). Individual tiers move to Antigravity	Developers in IDEs
NotebookLM	Source-grounded research assistant (summaries, Audio/Video Overviews, mind maps)	Researchers, students, teams

Rule of thumb:

Just trying it? → AI Studio, then the Gemini Developer API.
Building a normal app? → Gemini Developer API. Google's own guidance: most developers should use the Developer API unless they need specific enterprise controls.
Production with governance, data residency, GCP integration? → Gemini Enterprise Agent Platform.
Want to self-host or fine-tune open weights? → Gemma.
Coding in an IDE? → Code Assist / Antigravity.
Research over your own documents? → NotebookLM.

2. Google AI Studio + the Gemini Developer API

AI Studio is the free, browser-based front door. You can test prompts, compare models, tune parameters, and generate an API key without writing code. The Gemini Developer API is the same models behind a simple API-key auth.

Best fit when you are:

learning, prototyping, or running a quick experiment,
building a small-to-mid app that does not yet need enterprise controls,
on the free tier and fine with reduced rate limits.

Watch-outs:

The free tier has reduced rate limits, and free-tier data may be used for product improvement (per the pricing page).
AI Studio is for prototyping; production hosting and governance live on the Enterprise Agent Platform (see section 6).

3. Gemini Enterprise Agent Platform (formerly Vertex AI)

The enterprise platform on Google Cloud. Same Gemini models, but with the controls a production or regulated deployment needs:

Google Cloud service-account auth instead of a raw API key,
multi-region GCP deployment with configurable data-residency options,
MLOps plus tooling to build, govern, and observe agents at scale,
Model Garden (200+ models, including third-party models such as Claude).

For a regulated EU context, the configurable data-residency is the key differentiator versus the free Developer API. An official migration guide exists for moving an app from the Developer API to the platform.

4. The other surfaces, briefly

Gemini in Workspace — Gemini inside Gmail, Docs, Sheets, Slides, and Meet; for end users who want help where they already work.
Gemini Code Assist / Google Antigravity — IDE coding help (chat + agent mode, multi-file edits, MCP) for VS Code, JetBrains, and Cloud Shell. Note the transition: individual / AI Pro / Ultra tiers move to Google Antigravity (an agentic development platform + CLI) from June 18, 2026; Standard and Enterprise tiers remain on Code Assist.
NotebookLM — a source-grounded research assistant: upload documents, get cited summaries, Audio/Video Overviews, and mind maps. Included in Workspace plans.

5. Model lineup (current, mid-2026)

The family has moved to Gemini 3.x, with the 2.5 line still generally available.

Gemini 3.x (current generation):

Model	Status	Positioned for
`gemini-3.5-flash`	GA	Default model; sustained frontier performance on agentic and coding tasks
`gemini-3.1-pro-preview`	Preview	Advanced reasoning, complex problem-solving, agentic coding
`gemini-3.1-flash-lite`	GA	Most cost-effective Gemini 3-series model for high-volume work

Gemini 2.5 (still GA): gemini-2.5-pro (deep reasoning + coding; 1M-token context, officially stated), gemini-2.5-flash (best price/performance), gemini-2.5-flash-lite.

Context windows for 3.x are unverified here

The models overview page does not list exact per-model token limits for the 3.x line; only Gemini 2.5 Pro = 1M tokens is confirmed. Check each model's detail page before quoting 3.x context sizes.

Adjacent models:

Gemma 4 (open weights, Apache 2.0, commercial use allowed) — Google's open models, downloadable via Hugging Face, Kaggle, and Ollama; natively multimodal. This is the surface for self-hosting and fine-tuning.
Image / video — native image generation (the "Nano Banana" image models) and Veo for video generation.

Gemini models also expose adaptive thinking with controllable "thinking budgets."

6. Quickstart (zero → first API call)

Get a free API key at aistudio.google.com/apikey.
Install the SDK — the package is google-genai:

pip install -U google-genai
export GEMINI_API_KEY="your-key"

First call:

from google import genai

client = genai.Client()  # reads GEMINI_API_KEY from the environment

response = client.models.generate_content(
    model="gemini-2.5-flash",          # swap in gemini-3.5-flash for the current-gen default
    contents="Summarize this text in two sentences: ...",
)
print(response.text)

On the SDK surface

Google now recommends a newer Interactions API for access to the latest models and features; the classic generate_content method shown above is still supported. The exact request/response shape has changed recently, so treat code samples as a starting point and confirm against the official quickstart.

7. Decision guide

If you want to…	Use…
try models and get an API key for free	Google AI Studio
build a normal app ("most developers")	Gemini Developer API
run production with governance, data residency, and GCP integration	Gemini Enterprise Agent Platform
self-host or fine-tune open weights	Gemma
code inside an IDE	Gemini Code Assist / Google Antigravity
research over your own documents	NotebookLM
give business users AI inside Google apps	Gemini in Workspace or the Gemini app

8. Recommended starting points

Solo developer / startup: AI Studio → Gemini Developer API.
Production team with compliance needs: prototype on the Developer API, deploy on the Enterprise Agent Platform for data-residency and governance.
On-prem / privacy-sensitive: Gemma open weights (e.g. via Ollama).
Knowledge worker: the Gemini app or Gemini in Workspace; NotebookLM for document research.

9. Official links

Docs & developer

Products

Related guides

1. The mental model​

2. Google AI Studio + the Gemini Developer API​

3. Gemini Enterprise Agent Platform (formerly Vertex AI)​

4. The other surfaces, briefly​

5. Model lineup (current, mid-2026)​

6. Quickstart (zero → first API call)​

7. Decision guide​

8. Recommended starting points​

9. Official links​