Google I/O 2026 AI Launches: Gemini 3.5, Antigravity, Omni

Google I/O 2026 produced a long list of AI announcements across models, Search, Workspace, developer tools, subscriptions and DeepMind product/research updates. Treating that whole pile as one item in a directory would miss what is actually shipping. Several of the launches belong in different Toolhalla categories, are aimed at different buyers, and answer different questions.

This article maps the keynote's AI launches to the buying and building decisions Toolhalla readers actually face: which item is a model, which is a development platform, which is a multimodal product, and which is a productivity surface. We have not tested any of the announced products hands-on, so nothing here should be read as a benchmark or quality verdict. Where Google has published numbers, the numbers are theirs.

Short answer: one keynote, several different product categories

Google's "all our announcements" page bundles the I/O 2026 AI updates under a single hub, but the individual announcements span at least four distinct categories that Toolhalla already tracks separately:

• A model family. Gemini 3.5, introduced as "our latest family of models combining frontier intelligence with action," with 3.5 Flash described as Google's "strongest agentic and coding model yet."
• A multimodal generation model. Gemini Omni, positioned by Google as "a model that can create anything from any input — starting with video."
• A world-model research direction. Project Genie combined with Street View, framed by DeepMind as a "general-purpose world model capable of generating diverse, interactive environments," now anchored to real Street View imagery.
• A coding/agent development platform. Google Antigravity 2.0, described by Google as an agent-first development platform with a 2.0 desktop app, CLI and SDK, and multi-agent orchestration co-optimized with Gemini 3.5 Flash.

Search, Workspace, AI Studio and the new subscription tiers wrap around those four anchors. For a directory, the implication is that the keynote is not a single entry; it is updates to several different listings.

Gemini 3.5 belongs in the model directory, not the app directory

Google's Gemini 3.5 post introduces 3.5 as a model family rather than an app. 3.5 Flash is "available today" through the Gemini app, AI Mode in Google Search, Google Antigravity, the Gemini API, Android Studio, the Gemini Enterprise Agent Platform, and Gemini Enterprise. Google says 3.5 Pro is in internal testing and is set to roll out "next month."

Google's post also publishes benchmark claims for 3.5 Flash: Terminal-Bench 2.1 at 76.2%, GDPval-AA at 1656 Elo, and MCP Atlas at 83.6%, while running "4 times faster than other frontier models" in output tokens per second. These are Google's stated numbers from the announcement, not Toolhalla measurements. They are useful for filtering candidates; they are not a substitute for evaluating against your own prompts and traffic.

For Toolhalla's directory, the cleanest update is to track Gemini 3.5 the way other model families are tracked — provider Google, model family Gemini 3.5, variants Flash (GA) and Pro (announced, rolling out), distribution surfaces Gemini API, AI Studio, Antigravity, Android Studio, and Gemini Enterprise. Anyone who has been comparing assistants through our ChatGPT vs Claude vs Gemini for coding breakdown should treat 3.5 Flash as the new Google entry to evaluate, not as a separate product.

Gemini Omni and Project Genie point to multimodal and world-model workflows

Gemini Omni is a different shape of update. Google describes it as "a model that can create anything from any input — starting with video," with images, audio, video and text supported as inputs and video as the current output modality. Image and audio outputs are described as future support. The first variant, Gemini Omni Flash, is rolling out to Google AI Plus, Pro and Ultra subscribers via the Gemini app and Google Flow, and to YouTube users via Shorts Remix at no cost (18+).

Project Genie with Street View is adjacent but distinct. The DeepMind announcement calls Genie "our general-purpose world model capable of generating diverse, interactive environments," and the Street View integration lets users anchor those environments to real places. Google references applications in autonomous-vehicle simulation (citing Waymo) and in providing virtual environments for AI agents or robots to navigate. The Street View prototype is being rolled out to Google AI Ultra subscribers in the U.S. first.

For Toolhalla, the two updates belong in different buckets:

• Gemini Omni is a multimodal generation product to compare against other video-and-multimodal tools, with the caveat that input/output coverage is narrower than the name implies for now.
• Project Genie is a world-model research direction relevant to anyone tracking simulation, robotics, and agent-training environments. It is not a general-purpose creative tool, and it should not be listed as one.

Neither one is a substitute for Gemini 3.5 in the model directory; they sit alongside it.

Antigravity 2.0 is the coding-agent item to compare against Cursor, Codex and Copilot

Google Antigravity 2.0 is the announcement most likely to land on Toolhalla's AI coding lists. The Google I/O 2026 announcements page describes it as an agent-first development platform with a 2.0 desktop app, an Antigravity CLI and SDK for terminal and programmatic access, multi-agent orchestration, and co-optimization with Gemini 3.5 Flash. Google AI Studio's native Android app building path also references export to Antigravity for local development.

The category fit is clear: it is a coding-agent platform aimed at developers who want to drive multiple agents from one surface, with first-class CLI and SDK access. That is the same category Toolhalla readers already weigh in our comparisons of Claude Code, Cursor and GitHub Copilot and the best AI coding assistants in 2026. The buyer questions are familiar:

• Does the agent platform integrate with your existing repos, CI and review workflow?
• Is the CLI useful enough to drive without the desktop app?
• How does the multi-agent orchestration handle scope, permissions, and audit trails?
• Which model can it run with, and at what cost?

What Google's announcement does not establish is that Antigravity 2.0 is better than Cursor, OpenAI Codex, or GitHub Copilot. None of those head-to-head comparisons are in Google's post. Buyers should treat Antigravity 2.0 as a serious entrant in the coding-agent category and evaluate it directly against their current stack, not as a declared category leader.

Search, Workspace and AI Studio updates matter for buyer workflows

The keynote also bundles a long set of product updates that change how Google's surfaces are used day to day:

• Search. Google says Gemini 3.5 Flash is now the default model in AI Mode globally, that the Search box has been redesigned to accept multimodal inputs (text, images, files, videos, Chrome tabs), and that Search Agents will monitor topics on a 24/7 basis for Google AI Pro and Ultra subscribers from summer 2026. Personal Intelligence is described as expanding to nearly 200 countries and 98 languages without requiring a subscription.
• Workspace. Gmail Live, Docs Live and Talk to Keep add voice-driven workflows for inbox search, drafting, and note-taking. AI Inbox is rolling out to Google AI Plus and Pro subscribers with prioritized surfacing and draft replies. Google Pics adds image creation and editing inside Workspace. Gemini Spark is positioned as a 24/7 personal AI agent for Ultra subscribers in the U.S.
• AI Studio and developer building. Google AI Studio now supports native Android app building with Google Play Console integration, including export to Antigravity for local development. A Managed Agents API provisions a Linux environment per call for reasoning, code execution and web browsing, powered by Gemini 3.5 Flash via the Interactions API.
• Subscriptions. Google AI Ultra is now a $100 plan with described 5× usage limits and 20TB of storage. AI Pro now includes YouTube Premium Lite at no extra charge.

For Toolhalla, these updates change which products are worth re-evaluating in buyer workflows rather than creating new directory entries on their own. The voice features change the Workspace evaluation; Search Agents change the AI Mode evaluation; AI Studio's Android path changes the Android-builder evaluation. None of them turn Search or Workspace into a different product category.

What not to overclaim yet

Several things in the keynote are positioning, not proof. A short list of things worth not overclaiming:

• Benchmark scores. Terminal-Bench 2.1 76.2%, GDPval-AA 1656 Elo, MCP Atlas 83.6%, and "4× faster" output tokens per second are Google's stated numbers for Gemini 3.5 Flash. They are not independent measurements, and they do not necessarily reflect your prompts, your tools, or your evaluation harness.
• "Strongest" or "first" claims. Google calls 3.5 Flash its "strongest agentic and coding model yet" and Gemini Omni "a model that can create anything from any input." Those are Google's own framings. They are not head-to-head comparisons with other vendors' models or with Cursor, OpenAI Codex, GitHub Copilot, or other AI coding agents.
• Availability windows. Several features are described as rolling out over "summer 2026," to specific subscription tiers, in specific countries, or to specific age groups. Treat "announced" and "available to your team today" as different things.
• Gemini Omni's modality coverage. Inputs include images, audio, video and text. Outputs currently center on video. Calling Omni a fully omni-modal generator today would overstate what Google's own post claims.
• Project Genie's reach. It is a research-anchored world model accessible through an experimental prototype on Google AI Ultra. It is not a general-purpose creative tool yet, and it should not be evaluated as one.

The combination of those points is also why the keynote should not be flattened into a single directory entry: different items will mature on different timelines.

Toolhalla directory update checklist

Treating the I/O 2026 announcements as a list of directory work rather than a single post, the cleanest pass looks like this:

1. Models. Add or update Gemini 3.5 Flash (GA) and note Gemini 3.5 Pro as announced and rolling out. Keep Google's benchmark numbers attributed to Google.

2. Coding agents. Add or update Google Antigravity 2.0 as an agent-first development platform with CLI/SDK and multi-agent orchestration. Compare in the coding-agent category with Cursor, Codex and Copilot using buyer questions, not Google's framing.

3. Multimodal. Add Gemini Omni and Omni Flash as a multimodal generation model with video output today and image/audio outputs described as future. Note distribution via the Gemini app, Google Flow, and YouTube Shorts Remix.

4. World models / simulation. Add Project Genie with Street View as a world-model research direction, with availability through Google AI Ultra in the U.S. first. Do not list it as a creative tool.

5. Search. Update the AI Mode entry to reflect Gemini 3.5 Flash as the default model, the new multimodal Search box, and Search Agents as a subscriber feature.

6. Workspace. Update voice-driven Workspace entries (Gmail Live, Docs Live, Keep) and add Google Pics. Treat AI Inbox as a subscriber expansion, not a new product.

7. Developer building. Update AI Studio to reflect native Android app building and Antigravity export. Track the Managed Agents API separately as an agent-runtime primitive.

8. Subscriptions and bundles. Note the new $100 Google AI Ultra plan and the AI Pro / YouTube Premium Lite bundling so price-comparison entries stay accurate.

For broader context on how Gemini compares as a real-time and voice surface, see our earlier write-up on Gemini 3.1 Flash Live.

FAQ

What was the biggest AI announcement at Google I/O 2026?

There is no single "biggest" item. Google's own keynote post and the "all our announcements" page surface Gemini 3.5 (model family), Gemini Omni (multimodal generation), Project Genie with Street View (world model), Google Antigravity 2.0 (coding-agent platform), Search and Workspace updates, and new AI subscription tiers. Calling any one of them "the" headline collapses categories that Google itself treats separately.

Is Gemini 3.5 an AI tool or an AI model?

Gemini 3.5 is an AI model family. Google's post introduces it as "our latest family of models combining frontier intelligence with action," with 3.5 Flash generally available and 3.5 Pro described as rolling out the following month. The Gemini app, AI Mode in Search, Antigravity, the Gemini API, Android Studio, and Gemini Enterprise are surfaces that use Gemini 3.5 — they are not the model itself.

What is Google Antigravity 2.0 used for?

Google describes Antigravity as an agent-first development platform. The 2.0 release adds a desktop app, an Antigravity CLI and SDK, and multi-agent orchestration co-optimized with Gemini 3.5 Flash. It is positioned for developers who want to drive coding and agent work from a dedicated platform rather than a general chat product, and it shares category space with other AI coding agents.

How should buyers compare Gemini Omni with other multimodal AI tools?

Treat Gemini Omni as a multimodal generation model that currently outputs video, with image and audio outputs described by Google as future support. The first variant, Gemini Omni Flash, is distributed through the Gemini app, Google Flow, and YouTube Shorts Remix. Compare on the modalities you actually need today, on the surfaces you plan to use, and on the evaluation criteria you would apply to any other generation model — not on the "omni" name alone.

Sources

• Google blog, "Google I/O 2026: all our announcements": https://blog.google/innovation-and-ai/technology/ai/google-io-2026-all-our-announcements/
• Sundar Pichai, "Google I/O 2026 keynote": https://blog.google/innovation-and-ai/sundar-pichai-io-2026/
• Google blog, "Introducing Gemini 3.5": https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/
• Google blog, "Search at I/O 2026": https://blog.google/products-and-platforms/products/search/search-io-2026/
• Google Workspace updates blog: https://blog.google/products-and-platforms/products/workspace/workspace-updates/
• Google DeepMind, "Introducing Gemini Omni": https://deepmind.google/blog/introducing-gemini-omni/
• Google DeepMind, "Simulate real-world places with Project Genie and Street View": https://deepmind.google/blog/simulate-real-world-places-with-project-genie-and-street-view/
• Google DeepMind, "Introducing Google Antigravity 2.0": https://deepmind.google/blog/introducing-google-antigravity-2-0/