Google Gemini

Gemini is Google's family of AI models, offered through an API, in the Gemini app and inside Google Workspace apps, alongside Gemma, a family of open models from Google DeepMind.

For developers, the Gemini Developer API and Google AI Studio are the main path. According to the documentation, AI Studio is for testing prompts, managing API keys, monitoring usage and building prototypes.

For companies that need specific controls, there is the Google Cloud platform, which the documentation now calls Gemini Enterprise Agent Platform and which used to be called Vertex AI. It offers data residency features. The same Google Gen AI SDK calls both backends and, with a few exceptions, code that runs on one runs on the other.

For everyday use, Gemini appears in a side panel in Gmail, Docs, Sheets, Slides, Drive and Chat. The Gemini app, which also has a Windows version, connects to Gmail and Drive.

Gemini is aimed at developers who integrate AI models through an API, companies that use Google Workspace and Google Cloud, and teams that want to run open Gemma models locally.

Profiles and uses the documentation supports:

• Developers and prototypes: the Gemini Developer API with AI Studio. The documentation recommends the Developer API unless you need specific enterprise controls.
• Companies with governance requirements: the Google Cloud platform, with data residency features.
• Workspace users: a side panel to summarize threads and draft emails in Gmail and documents in Docs, and Deep Research, which can use Drive, Gmail, Chat and the web as sources.
• Realtime voice applications: the Live API, with function calling and Google Search.
• Teams that need a local model: the Gemma family from Google DeepMind. The Gemma 3n page describes it as designed to run on phones, tablets and laptops.

Before starting, access to the Developer API uses an API key managed in AI Studio. For Gemma, you need hardware that fits the size you choose, and the listed sizes range from mobile and IoT models to larger variants.

Gemini's main features are the models available through the API, Google Search grounding, the Live API for voice, native image generation and Workspace integration.

• Google Search grounding: when you turn on the google_search tool, the model searches, processes and cites sources automatically.
• Other built-in tools: Google Maps, code execution, URL context, computer use (in preview) and File Search, which indexes and searches your own documents for RAG.
• Function calling: according to the Gemini 3 guide in the documentation, those models can combine built-in tools and your own functions in one call.
• Live API: realtime conversations with function calling and Google Search. Synchronous or asynchronous function execution varies by model.
• Images: native generation and editing from text prompts.
• Deep Research: reports that can use Gmail, Drive, Chat and the web as sources.
• Workspace side panel: thread summaries and email drafting in Gmail, writing help in Docs, and use in Sheets, Slides, Drive and Chat.
• Gemma: Google DeepMind's open family, multimodal, with text and image input and text output.

Gemini is worth it for teams that already live in Google Workspace or Google Cloud and want models through an API, in the app and in an open version, with the caveat that part of the features are in preview.

Pros:

• There are two doors to the same API: the Developer API to get started and the Google Cloud platform for enterprise controls, with the same SDK.
• Google Search grounding has the model itself search, process and cite sources.
• Gemini is inside Gmail, Docs, Sheets, Slides, Drive and Chat, where many teams already work.
• The Gemma family gives an open-model path, including sizes for phones and IoT.

Cons:

• There are two surfaces to choose between, and code runs on both only with a few exceptions, according to the documentation.
• Some features appear as preview, such as computer use, and may change.
• The Google Cloud platform was renamed (formerly Vertex AI), so older tutorials use the previous name.
• In the Live API, synchronous or asynchronous function calling depends on the model, so you have to check each model's page.

The main alternatives to Gemini are Anthropic, OpenAI, Meta AI and SpaceXAI, labs that also offer language models through APIs and apps.

• Anthropic: offers the Claude family through the Claude API, apps and Claude Code, and also through Google Cloud, Amazon Bedrock and Microsoft Foundry.
• OpenAI: offers the Responses API, ChatGPT and Codex, and publishes the open-weight gpt-oss models.
• Meta AI: publishes open-weight models, such as the Llama family and Muse Spark Open, and hosts other models on the Meta Model API.
• SpaceXAI: the Grok models' API includes web search, X search and code execution as built-in tools.

Anyone looking for an open model like Gemma has options in OpenAI (gpt-oss) and Meta. Anyone on Google Cloud can have Anthropic in the same cloud, according to Anthropic's documentation.