What are mini-apps in Google AI Mode?

Mini-apps are interactive, single-purpose web apps that Google AI Mode generates on demand from a natural-language description. Announced at Google I/O 2026 on May 19, 2026, they're built by Antigravity 2.0 (Google's agentic IDE) running server-side inside AI Mode. You ask AI Mode 'build me a tool to compare three mortgage scenarios with my numbers,' and it produces a working interactive app — not just a static answer. Mini-apps are launching first to US Google AI Pro and AI Ultra subscribers.

What is generative UI in Google Search?

Generative UI is the underlying capability that lets AI Mode build custom interfaces per query — interactive 3D models for explaining concepts, personalized dashboards, decision trees, calculators, comparison tables with sliders, and more. Unlike mini-apps (which are user-initiated and saveable), generative UI is automatic — AI Mode decides when a custom interface will explain something better than text. Generative UI is free for all AI Mode users globally.

What's the difference between generative UI and mini-apps?

Generative UI is automatic and ephemeral — AI Mode generates a custom UI for one query, then it's gone. Mini-apps are explicit and persistent — you ask AI Mode to build a tool, save it, share it, and reuse it later. Generative UI is free worldwide; mini-apps require AI Pro or AI Ultra and roll out to the US first. Both use the same underlying Antigravity 2.0 + Gemini 3.5 Flash stack.

How does Google generative UI compare to Claude Artifacts and ChatGPT Canvas?

Claude Artifacts (Anthropic) and ChatGPT Canvas (OpenAI) are chat-only — they live inside the chat UI and render simple React/HTML artifacts. Google's generative UI lives inside Search itself, which means hundreds of millions of users encounter it without opting into a chat app. It also uses Antigravity 2.0's agent harness, which can call external tools and APIs to populate the UI, not just render static code. Tradeoff: Artifacts and Canvas give you more granular edit control; Google's generative UI is more 'magic, less control.'

Quick Answer

What Are Google Mini-Apps & Generative UI? (May 2026)

Published: May 22, 2026

What Are Google Mini-Apps & Generative UI? (May 2026)

Google AI Mode now builds interactive apps and custom UIs on the fly, powered by Antigravity 2.0. Announced at Google I/O 2026 on May 19, 2026, this is the most ambitious shift in how Search renders answers since the Knowledge Graph.

Last verified: May 22, 2026

Quick facts

Feature	Generative UI	Mini-apps
Announced	May 19, 2026	May 19, 2026
Surface	AI Mode (Google Search)	AI Mode (Google Search)
Powered by	Antigravity 2.0 + Gemini 3.5 Flash	Antigravity 2.0 + Gemini 3.5 Flash
Trigger	Automatic (AI decides)	User-initiated (“build me a…”)
Persistence	Ephemeral (per query)	Saved + shareable
Pricing	Free globally	Google AI Pro / Ultra (US first)
Best for	Explanations, exploration	Tools, calculators, comparisons

What generative UI is

Traditional Google Search shows you links and AI Overviews show you paragraphs. Generative UI does something different: when AI Mode decides a custom interactive interface would explain your query better than words, it builds one.

Examples Google showed at I/O 2026:

“Explain how a transformer model works” → Interactive 3D diagram with attention heads you can click and inspect.
“Compare these three apartments” → Side-by-side dashboard with map, photos, score-cards, and a “weight what matters” slider panel.
“Show me how interest rates affect my mortgage” → Live calculator with sliders for rate, term, principal, and a payment-over-time chart.
“Walk me through baking sourdough” → Step timeline with timers, temperature alerts, and embedded video at each step.

These aren’t pre-built UIs from a template library. AI Mode literally generates the HTML/JS/CSS per query using Antigravity 2.0’s agent harness running server-side at Google.

Pricing: Generative UI is free for all AI Mode users in 200+ countries.

What mini-apps are

Mini-apps are the explicit, user-initiated, persistent version of generative UI. Instead of waiting for AI Mode to decide a query deserves a custom interface, you ask for one:

“Build me a tool that compares three mortgage scenarios using my numbers.”

“Make me a workout tracker for a 12-week strength program.”

“Create a flashcard app for the 200 Mandarin characters I’m studying.”

AI Mode generates a working app, you can use it, save it, and come back to it. You can also share it via link with anyone (recipients don’t need an AI Pro / Ultra subscription to use a shared mini-app, only to create new ones).

Pricing: Mini-app creation is gated to Google AI Pro ($19.99/mo) and Google AI Ultra ($99.99 / $199.99 /mo) subscribers, US first. Other markets follow over the next quarter.

Why generative UI and mini-apps now

Three things had to happen first:

Antigravity 2.0 — Google’s agent harness needed to be capable of generating, running, and sandboxing arbitrary frontend code reliably. Antigravity 2.0 shipped at I/O 2026.
Gemini 3.5 Flash — Fast enough at low enough cost to do this for billions of Search queries. Gemini 3.5 Flash also shipped at I/O 2026 and is the default model behind both features.
Sandboxing infrastructure — Running untrusted (model-generated) code at Google Search scale requires the kind of isolation Google already built for Chrome and Cloud Run.

The result: a Search box that doesn’t just answer questions, it builds the tool that answers them.

Generative UI vs Claude Artifacts vs ChatGPT Canvas

	Google generative UI	Claude Artifacts	ChatGPT Canvas
Vendor	Google	Anthropic	OpenAI
Surface	Google Search / AI Mode	Claude.ai / API	ChatGPT app
Reach	Billions of Search users	Claude users only	ChatGPT users only
Auto-trigger	Yes	Sometimes	No (Canvas is opt-in)
Saveable / shareable	Mini-apps yes, gen UI no	Yes	Yes
External tool calls	Yes (Antigravity harness)	Yes (Claude tools)	Limited
Editable by user	Limited	Full	Full inline
Languages	JS/HTML/CSS (web)	JS/HTML/CSS, React, SVG, code	JS/HTML/CSS, Python, code
Free tier	Yes (gen UI)	Yes (limited)	Plus required for full Canvas

Each vendor is solving the same idea — generative interfaces inside a chat or search experience — at different scales. Google’s bet is the largest: Search reaches billions; Anthropic and OpenAI each reach hundreds of millions of paying / engaged users.

Mini-app examples that actually shipped at I/O 2026

Google’s keynote highlighted demos in these categories:

Personal finance — Mortgage comparators, retirement scenario builders, expense analyzers.
Education / learning — Flashcard apps, quiz builders, interactive course outlines.
Productivity — Project trackers, decision matrices, meeting agenda builders.
Health & fitness — Workout planners, macro calculators, supplement timers.
Travel — Itinerary builders, packing-list generators, multi-stop fare comparators.
Hobby / collector — Card collection trackers, wine cellars, watch comparators.

The pattern: single-purpose, opinionated tools that would normally cost a developer-day to build and that users would pay $5 / month for from a SaaS. Now generated for free in 30 seconds inside Search.

Limitations as of May 22, 2026

No third-party API auth — Mini-apps can call public APIs but can’t yet OAuth into your accounts (Gmail, Stripe, Notion, etc.).
No data persistence at scale — Mini-app state is per-user and per-app, no shared databases.
Mobile-first UIs are best — Complex desktop apps degrade.
English-first — Generative UI works in 98 languages; mini-app creation prompts work best in English currently.
No embed elsewhere — Mini-apps live at a Google URL; no <iframe> embedding into your site as of May 22, 2026.
Audit log / safety constraints — Mini-apps run inside Google’s sandbox; they can’t access local files, devices, or network beyond a curated allowlist.

What this means for builders, content creators, and SaaS founders

Builders. If you’ve been building a “calculator + form + chart” SaaS, the floor of “is this worth paying for” just went up. Single-purpose tools are now a Google query away.

Content creators. Pages that previously got long-tail traffic for “how do I calculate X” or “compare Y” queries will increasingly lose those clicks to in-Search generative UI. Anchor your content in opinion, framework, and proprietary data — things AI Mode can’t synthesize from public sources.

SaaS founders. Defensibility shifts from UI to data, integrations, workflow, and trust. A custom mortgage calculator isn’t defensible; a calculator wired into your bank, credit reports, and a licensed broker is.

TL;DR

Generative UI = AI Mode automatically builds a custom interactive interface for your query. Free, global, ephemeral.
Mini-apps = You explicitly ask AI Mode to build a saveable, shareable tool. AI Pro / Ultra, US first.
Both powered by Antigravity 2.0 + Gemini 3.5 Flash, shipped at Google I/O 2026 (May 19, 2026).
Closest competitors: Claude Artifacts (Anthropic), ChatGPT Canvas (OpenAI) — both chat-only, smaller reach.
Limits: no third-party OAuth, English-best, no <iframe> embed.

Sources

Google blog: “Search’s I/O 2026 updates” (May 19, 2026)
The Next Web: “Google replaces the search box with AI agents at I/O 2026” (May 20, 2026)
PCMag: “Omni, Pics, and a big Search overhaul: everything you missed at Google I/O 2026” (May 20, 2026)
Engadget: “Everything Google announced at I/O 2026” (May 19, 2026)