Google I/O 2026 Recap: Gemini 3.5 Flash, Omni Video, Spark Agent, Search Upgrade

Gemini 3.5 Flash

Gemini 3.5 Flash is positioned as a next‑generation model that combines frontier intelligence with strong agent capabilities. It outperforms Gemini 3.1 Pro on almost every benchmark, with especially large gains in coding tasks and strong results on the GDPVal metric for complex real‑world economic activities.

Speed – four times faster than peer frontier models; token‑per‑second throughput exceeds GPT‑5.5 and Claude Opus 4.7.

Cost efficiency – execution cost is less than half of other frontier models. Google estimates that shifting 80 % of a 10 trillion‑token daily workload to Flash could save >$1 billion per year.

Benchmarks – Terminal‑Bench 2.1, GDPval‑AA, and MCP Atlas all show Flash surpassing 3.1 Pro in coding and agent evaluations.

Current status : Flash is fully open today via the Gemini app, Google AI Studio, and API, and serves as the default model for AI Mode in Search. Gemini 3.5 Pro is slated for release next month.

Gemini Omni

Gemini Omni is a multimodal model that supports arbitrary input and video output. The first batch of video generation is available.

Generate new videos from text, images, or existing video material.

Edit uploaded videos – change scenes, add objects, apply styles.

Define a character and maintain its consistency across scenes.

Apply styles, motion, or effects via natural‑language prompts.

Community experiments include turning people in videos into flamingos, visualizing photosynthesis, and creating cinematic‑grade effects.

Current status : Gemini Omni Flash is available to Google AI Plus, Pro, and Ultra subscribers; API access will open in the coming weeks.

Gemini Spark

Gemini Spark is a dedicated AI agent within the Gemini app that can operate continuously without user supervision.

Runs on a dedicated Google Cloud VM, operating even when the user’s laptop is off.

Powered by Gemini 3.5 plus the Antigravity framework, handling long‑running, multi‑step tasks.

Integrates tightly with Gmail, Docs, Slides; future third‑party integration via the MCP protocol.

Interaction modes: direct conversation in the Gemini app, email, or instant‑messaging.

Android Halo UI (launching later this year) provides a persistent space to monitor Spark tasks.

Chrome integration planned for summer, turning the browser into an “agent‑style” experience.

Current status : Spark began rolling out to trusted testers this week and will be available to Google AI Ultra subscribers in the U.S. next week as a beta.

Search Upgrade

AI Search Box

The traditional search box is replaced by an AI‑enhanced box that accepts text, images, files, and video, understands complex intent, and supports follow‑up questions. AI Overview and AI Mode are merged into a unified experience.

Information Agents

Users can create personalized AI agents that run 24/7, continuously searching, filtering, and pushing relevant information, and can even take actions on the user's behalf.

Mini‑Apps in Search

Built on Gemini 3.5 Flash and Antigravity, search can generate custom interactive experiences—visual simulations, dashboards, progress trackers—tailored to specific tasks.

Current status : The new AI search experience is live globally on desktop and mobile; information agents and mini‑apps will roll out throughout the summer.

Antigravity 2.0

Antigravity evolves from a programming environment into a platform for developing and managing multiple autonomous AI agents.

Antigravity 2.0 Desktop App – central hub for agent interaction; multiple agents can collaborate.

Antigravity CLI – terminal interface for rapid agent deployment without a GUI.

Antigravity SDK – toolkit for integrating the agent framework directly into software.

When paired with an optimized version of Flash, Antigravity runs up to 12 times faster than other frontier models.

TPU 8

TPU 8t (training) – raw compute power is three times that of the previous generation; leverages JAX + Pathways for cross‑data‑center distributed training, forming a global cluster of over one million TPUs.

TPU 8i (inference) – significantly accelerates each inference step, continuing Google’s focus on low latency.

Both chips deliver twice the performance per watt.

Google AI Subscription Plans

$100/month AI Pro – targets professional users needing more AI capability.

AI Ultra reduced to $200/month – premium tier price cut, includes full AI feature set and priority access.

Additional Notable Updates

Google Flow + Gemini Omni

Google Flow introduces a new AI agent for complex task planning, reasoning, and creation. It supports batch editing, role consistency, and a “Flow Tools” feature that lets users craft tools via natural‑language “vibe code”.

Daily Brief

A Gemini Agent that aggregates email, calendar, and task data each morning, ranks items by importance, and suggests next steps.

Ask YouTube

Reimagines YouTube search, matching videos to user interests and jumping directly to relevant segments.

Docs Live

Voice‑driven document creation: spoken ideas are automatically structured into documents, initially for subscribers and later for Gmail and Keep.

SynthID + Content Credentials

AI content transparency initiative: partners OpenAI, Kakao, ElevenLabs join the SynthID watermark system; Gemini app, Search, and Chrome integrate C2PA Content Credentials verification.

Google Pics

New AI image creation and editing tool based on the Nano Banana model, allowing independent manipulation of each element.

Smart Glasses Partnerships

Collaborations with Samsung, Gentle Monster, and Warby Parker to launch smart‑glass products later this year.

Workspace Integration in AI Studio

Developers can connect Docs, Calendar, and other Workspace apps to AI Studio with a single prompt.

Android Native Development Support in AI Studio

AI Studio now supports native Android app development; one prompt can generate a high‑quality Android app, with one‑click export to Antigravity for seamless prototyping to production.

Modern Web Guidance

Early preview of guidance for AI agents to build modern web features.

Universal Cart

Feature slated for summer to provide a unified shopping‑cart experience across Search, Gemini app, YouTube, and Gmail.

Build with Gemini XPRIZE

Hackathon announced with a $2 million prize to encourage developers to build real‑world solutions using Google’s agent tools.

Gemini for Science

Integration of Notebook LM, AlphaEvolve, and Co‑Scientist tools to aid scientists in literature insight, hypothesis generation, and computational discovery.

Key Metrics

Monthly processed tokens: >3,200 trillion (7× YoY growth)

AI Overview MAU: >2.5 billion

AI Mode MAU: >1 billion (1 year live)

Gemini app MAU: >0.9 billion (doubled in a year)

Developers per month: >8.5 million

Nano Banana generated images: >50 billion

Annual capital expenditure: ≈ $190 billion