AI
Gemini 3.5 Flash and Anti-Gravity 2.0
Gemini 3.5 Flash beats 3.1 Pro on almost all benchmarks and runs four times faster than other frontier models. Varun Mohan demonstrated Anti-Gravity 2.0, an agent-first desktop app. 93 sub-agents worked in parallel for 12 hours, making 15,000 model requests and processing 2.6 billion tokens. They built a functioning operating system from scratch: scheduler, memory management, file system. Cost: less than $1,000 in API credits. The OS even ran Doom after agents added video and keyboard drivers.
Gemini Spark: 24/7 Personal Agent
Gemini Spark runs on dedicated virtual machines on Google Cloud and works around the clock. Josh Woodward showed it drafting emails, managing block party RSVPs, and creating a Google Sheets tracker connected to Gmail. Spark syncs across devices; a voice command on a phone kicked off three tasks that continued in the background. It rolls out to trusted testers this week and to Ultra subscribers next week at $100 per month.
Search Agents and Generative UI
AI Search now includes information agents that monitor the web 24/7. Liz Reid demonstrated an agent tracking biotech stocks with P/E under 15, positive cash flow, and low debt. It connects to real-time finance data and sends synthesized updates. Robby Stein showed generative UI: search builds custom interactive visuals on the fly. A question about binary black holes produced a simulation with adjustable orbital separation and mass ratio. This uses the same Anti-Gravity coding harness that built the OS.
Agent E-Commerce: Universal Cart and Protocols
Three building blocks for agent e-commerce: Universal Commerce Protocol (UCP), Agent Payments Protocol (AP2), and Universal Cart. UCP is an open standard co-developed with Amazon, Meta, Microsoft, Salesforce, and Stripe. AP2 lets agents make purchases under user-set boundaries with tamper-proof digital mandates. Universal Cart works across merchants, tracks price drops, catches compatibility errors (e.g., wrong CPU socket), and finds hidden savings from Google Wallet offers. It rolls out in the US this summer.
Gemini Omni: From Any Input to Any Output
Gemini Omni Flash, available today in the Gemini app, generates video from text, images, or other video. Demis Hassabis showed a protein folding explainer created from a simple prompt. Josh Woodward demonstrated a singer transforming raw footage into a styled video with 360-degree camera angles. Omni understands physics like kinetic energy and gravity. It also edits videos conversationally: change the background, add elements, or adjust color. 50 billion images have been generated with Nano Banana models, and Omni extends that to video.
New Creative Tools: Pix, Stitch, and Flow
Google Pix, a new Workspace tool, removes objects, resizes, and translates text in images. Stitch generated over 100 million UI screens in the past year; now it builds websites from voice prompts. Google Flow added an agent that turns one image into 16 videos, custom tools via vibe coding, and music remixing. All outputs are watermarked with Synth ID, which has marked over 100 billion images and 60,000 years of audio. OpenAI, Kakao, and Lean Labs adopted Synth ID today.
Notable Quotes
Never imagined I’d say quadrillion in an IO keynote, but here we are. Sundar Pichai · ▶ Watch (4:32)
That never gets old. Varun Mohan · ▶ Watch (31:48)
AGI is now on the horizon, and it will be the most profound and impactful technology ever invented. Demis Hassabis · ▶ Watch (1:45:15)
We believe the best version of search is one created just for you. Robby Stein · ▶ Watch (50:48)
Key Takeaways
- Gemini 3.5 Flash delivers frontier intelligence at four times the speed of competitors.
- Anti-Gravity 2.0 sub-agents built a working OS in 12 hours for under $1,000.
- Gemini Spark is a persistent, 24/7 personal agent that takes action across apps.