Google Gemini

Gemini runs on a family of models in 2026. Gemini 3.5 Flash (launched at I/O 2026 on 19 May) is the new default across the Gemini app and Workspace. It combines frontier intelligence with agentic capability at 4x faster output than 3.1 Pro, and excels on coding and agentic benchmarks. Gemini 3.5 Thinking is the reasoning model, made for tasks that need step-by-step logic and longer deliberation. Gemini 3.1 Pro (February 2026) is a strong reasoning model for the most complex tasks, with a 1M token context window. Gemini 3.5 Flash is the default for everyday work. Gemini Omni Flash is an emerging video model that generates video from text, images, and clips. Gemini 3.5 Pro is in testing, expected "next month".

Workspace Intelligence (launched 22 April 2026) is the intelligence layer that maps your emails, files, chats, meetings, and collaborators across Gmail, Drive, Calendar, Chat, Docs, Sheets, and Slides into live context for Gemini. It works automatically. By default, Gemini has this context loaded before you write a prompt. The goal is to stop making workers feed context to the AI on every request. Third-party connectors (HubSpot, Salesforce, Asana, Jira) extend it beyond Google's own apps, and data is never used for ads or model training without permission.

Personal Intelligence connects the Gemini app to your Gmail, Drive, and Calendar to automatically prepare context before meetings and tasks, reading relevant email threads, documents, and calendar events without manual input. It is the personal counterpart to Workspace Intelligence. Workspace Intelligence covers your org context, Personal Intelligence covers your individual life and work patterns. It is available on Google AI Pro and Ultra personal subscriptions, and is currently for personal Google accounts. For Workspace enterprise users, the equivalent is Workspace Intelligence.

Gemini's custom instructions feature is called "Instructions for Gemini" and lives inside the Personal Intelligence settings (Settings, then Personal Intelligence, then Instructions for Gemini). You set them once and they apply to every chat, your role, communication style, and preferred output format. A few caveats your participants need to know. It requires a personal Google account and a Google AI plan, and is not available on work, school, or supervised Workspace accounts, so most business attendees on a Workspace login will not see it here. It does not apply to Gems or Live chats, and is unavailable in the EEA, Switzerland, and the UK (Australia is fine). For Workspace business users, the equivalent is persistent custom instructions set inside the Gemini side panel in Google Docs.

Deep Research is Gemini's autonomous research agent. It plans a research approach, searches hundreds of web sources, reads full articles, cross-references findings, and delivers a structured cited report. Before it runs, it presents a research plan you can review and modify. Reports run in 5 to 15 minutes depending on complexity, and you can upload your own files and include Workspace data alongside public web sources. Reports can become Audio Overviews, export to Google Docs, or convert into Canvas content. Limits vary by plan: Free gives up to 5 reports a month, AI Pro 20 a day, AI Ultra higher still. Deep Research Max (in preview for API and enterprise) adds collaborative planning, visualisations, MCP server connections, and File Search for accuracy-critical, long-running work.

Canvas is Gemini's side-by-side editing environment, a live document or code workspace that appears alongside the chat. You describe what you want, Gemini generates a first version in Canvas, and you refine through conversation and direct editing. Canvas integrates with the full Gemini toolset. A Deep Research report can become a Canvas document, a Canvas document can become an Audio Overview, and a Canvas doc can export directly to Google Docs. Available on all plans.

Gemini Live enables natural, real-time spoken conversation with Gemini, including the ability to interrupt, pause, and pick up where you left off. With the I/O 2026 redesign it is now an inline experience. You no longer have to switch to a separate fullscreen mode, it opens directly in the chat interface. Live supports Workspace Intelligence and Personal Intelligence context, so Gemini can reference your emails and files in voice conversations. Available on Android, iOS, and the web.

Daily Brief is a proactive agent that surfaces every morning as a personalised digest, pulling from your inbox, calendar, and task list to deliver a prioritised overview of the day ahead. It does not just summarise. It suggests next steps and surfaces the most pressing items first. It marks the shift of Gemini from reactive chatbot to proactive personal operating system. It launched on 19 May 2026 for US AI Plus, Pro, and Ultra subscribers, with international rollout and Workspace business access to follow.

Gemini generates images using Google's Nano Banana model (Gemini 3.1 Flash Image), reached via the "Create image" shortcut in the new tool bar below the prompt. Nano Banana's distinctive advantage is real-time Google Search grounding, so it can incorporate accurate visual reference (current geography, real buildings, verified product appearances) where other models would hallucinate. AI Pro and Ultra subscribers get higher-quality Nano Banana Pro. Google Pics (announced at I/O 2026) will become the dedicated image creation and editing app this summer, offering more precise control than prompt-only generation.

Gemini Omni Flash (launched at I/O 2026) is the first model in the Omni family, generating high-quality video from text prompts, uploaded images, and existing video clips. You edit via conversational prompts instead of traditional timelines: "remove the person in the background", "change the lighting to golden hour", "preserve the character across all scenes". It is available in the Gemini app (AI Plus, Pro, Ultra) and in Google Flow for creative video production, and maintains character consistency across scenes. The full Gemini Omni model, which also adds image and text output, is still in development.

Gems are custom AI assistants you build with a defined role, knowledge sources, and default tools. Unlike Instructions for Gemini, which apply globally, a Gem only activates when you open it, keeping your different AI contexts clean and separate. A Gem can include uploaded files, Google Drive connections, NotebookLM notebooks as knowledge sources, and a set of default tools (Deep Research, image generation, and others). Google provides pre-made Gems (coding assistant, writing editor, research analyst), and you can build unlimited custom ones. Gems are shareable, so you can hand a team member a Gem you built and it carries all the instructions and knowledge sources.

Gemini Spark is Google's answer to agentic AI, a 24/7 cloud-based personal agent built on Gemini 3.5 and the Antigravity platform. Spark is designed to work continuously in the background across Gmail, Docs, Slides, and partner apps via Model Context Protocol, proactively managing tasks and taking actions on your behalf. It asks for approval before sensitive actions like sending emails, adding calendar events, or making purchases. The US AI Ultra beta is rolling out from 26 May, and Workspace business customers get preview access this summer.

Deep Think is Google's reasoning mode for complex, open-ended problems that need deep logic, step-by-step planning, and rigorous self-verification. Instead of answering immediately, it uses reinforcement learning and parallel thinking to explore multiple hypotheses, poke holes in its own logic, and correct mistakes before it answers. It is built on Google's frontier models (Gemini 3.1 Pro) and made for deep work in science, mathematics, coding, and engineering. It is a premium feature. It is available to Google AI Ultra subscribers in the Gemini app via the "Thinking Level" toggle, and to select researchers and enterprises through the Gemini API.

Scheduled Actions let you set up tasks that Gemini executes at a specified time, recurring reminders, research queries, summaries, or any action you want automated on a schedule. It is the simpler, lighter version of agentic scheduling, available now on AI Pro plans ahead of the full Spark agent rollout. It is useful for recurring daily or weekly tasks where you want the output waiting for you rather than needing to initiate it.

NotebookLM is a grounded research workspace. You add sources (PDFs, Google Docs, Drive folders, YouTube links, websites, audio files, and now entire SharePoint sites via Workspace integration) and Gemini reasons only within those sources. It will not make things up beyond your material. From your sources it produces structured summaries, Audio Overviews (podcast-style summaries you can listen to), interactive Q&A, mind maps, study guides, flashcard sets, and written briefs. Notebooks can be shared with colleagues who can then chat with the same material. It is powered by Gemini 3.1 Pro for AI Pro and Ultra subscribers, with up to 300 sources per notebook on paid plans.

Audio Overview converts any set of uploaded sources, or a Deep Research report, Canvas document, or Gemini conversation, into a spoken audio track that sounds like a short podcast discussion. Two AI voices discuss the key themes, explain concepts, and highlight the most important points. It is available in the Gemini app and NotebookLM. It is particularly useful for consuming research during commutes, travel, or when switching from screen-based work to movement.

Gemini in Gmail drafts replies in your writing style, summarises long threads with citations, and auto-detects calendar invites buried in emails. "Help me write" generates first drafts from a brief. The "AI Overview" feature lets you ask questions about your email history ("Who gave me a quote for the Melbourne office fit-out?") and Gemini searches across your entire inbox to find it. Smart Compose suggests completions as you type. It is available on mobile (iOS and Android) as well as desktop, and implicit grounding means you can reference email threads in your prompts without manually attaching them.

"Help me create" in Docs generates a complete first draft pulling from Drive, Gmail, and Chat. You describe what you want, and Gemini assembles it with context from your workspace. Automatic summarisation surfaces key points from any long document, and citations are now displayed automatically whenever Gemini draws from web content or Workspace sources. Gemini can create fully styled, formatted documents, not just plain text, choosing appropriate layouts, headings, and visual structure. Voice features coming this summer will let you speak directly to Gemini inside Docs.

Gemini in Sheets can generate a fully structured spreadsheet from a single prompt, pulling data from Gmail, Drive, and Chat. "Fill with Gemini" speeds up populating tables from raw or incomplete data. It generates formulas and pivot tables in plain English, creates charts and visualisations, and analyses datasets for patterns and anomalies. Workspace Intelligence means it knows what project you are working on and can pull relevant numbers without you specifying the source.

Gemini in Slides suggests layouts, rewrites slide text for different audiences and tones, generates Nano Banana images directly for slides, and can build a full deck from a brief. Google Pics (coming this summer) will add more precise image editing directly inside Slides. Export from Canvas to Slides is one click. For internal presentations, Gemini's formatting capability alone, consistently applying layout and structure across a deck, saves significant manual work.

Gemini in Meet automatically transcribes meetings, generates structured summaries with key decisions and action items, and produces Audio Overviews you can listen to afterwards. "Take Notes with Gemini" runs in the background, so you do not need to activate it mid-meeting. After the meeting, you can ask Gemini questions about what was discussed from the transcript. Meeting notes can export to Docs, convert to Canvas, or go to a NotebookLM notebook, and live captions in 65+ languages are also powered by Gemini.

Workspace Studio (GA December 2025) is Google's no-code platform for building AI agents and workflow automations inside Google Workspace. Describe an automation in plain English and Workspace Studio builds it, connecting Gmail, Sheets, Drive, and third-party apps. Early examples include competitor research workflows that used to take hours, automated booking confirmations for service businesses, and content batch-processing for teams. It is the equivalent of Microsoft's Copilot Studio, built for the Google Workspace ecosystem.