Gemini 3.5 Flash and the Next Wave of Omni Models
At Google I/O, Sundar Pichai used the opening keynote to formally launch Gemini 3.5 Flash, a new high-speed model that had briefly surfaced under the “Gemini 3 Fast” label before the event. Positioned as the lightweight sibling in the Gemini family, 3.5 Flash is designed to power responsive, agentic experiences across Search, Antigravity, and other Google surfaces. Alongside Flash, Google highlighted the emerging Gemini Omni model as its next-generation multimodal backbone, bringing text, image, and video understanding into a single architecture. A dedicated Gemini Omni video capability was showcased in creative tools like Google Flow, where Omni-driven models help maintain character consistency and visual coherence across scenes. While Gemini 3.5 Pro remains “coming soon,” the broader message was clear: Gemini is evolving into a tiered stack of models optimised for speed, depth, and multimodal reasoning, ready to sit underneath almost every new feature unveiled on stage.
Unified AI Search: From AI Overviews to Agentic Answers
Google’s search experience is being reimagined around deeper AI search integration. The company is merging its AI Overviews and AI Mode concepts into a single, unified interface that lets users move fluidly between classic links and Gemini-powered answers. In demos, Generative UI in Search—driven by Antigravity and Gemini 3.5 Flash—planned responses “from scratch,” surfacing interactive, personalised layouts instead of static text snippets. A weekend itinerary example showed Search using existing context, like calendar entries and personal preferences, to shape results in a more actionable way. This unified experience also underpins new commerce features such as Universal Cart, an AI shopping agent that tracks products across Search, Gemini, YouTube, and Gmail while monitoring price history and stock. Together, these updates show Google treating search less as a list of pages and more as a conversational workspace where agents orchestrate planning, research, and purchasing in one continuous flow.
Spark, AI Studio, and New Tools for Agentic Developers
Developers were a key audience for Google’s Gemini announcements. The new Spark mode in the Gemini desktop app introduces an agent-style environment that can work directly with local folders, connectors, and user-defined skills, pushing Gemini toward more autonomous workflows. Google AI Studio, which previously lived on the web, is gaining a dedicated mobile app so developers can prototype prompts and code against Gemini models from their phones. On the web platform side, Google introduced Modern Web Guidance, new Chrome DevTools tailored for agents, and WebMCP, a framework that turns web pages into structured toolkits for Gemini-powered agents. These releases aim to make it easier to embed Gemini 3.5 Flash and the broader Gemini Omni model into production apps. Combined, they signal Google’s intent to make agentic behaviour—a system that can plan, call tools, and iterate—feel like a default capability rather than an advanced integration.
Ask YouTube, Gemini UI Refresh, and Android XR Integration
Beyond core models, Google spotlighted how Gemini is spreading into consumer apps and new hardware form factors. A standout addition is Ask YouTube, which lets users query videos directly, tapping Gemini to surface specific moments or explanations without manual scrubbing. The Gemini app itself is getting a “complete redesign” using a visual language called Neural Expressive, along with Gemini Live for rapid switching between typing and natural conversation. Responses are now chunked with graphics and imagery instead of appearing as monolithic text walls. On the platform side, Sundar Pichai also emphasised Android XR, showcasing new intelligent eyewear powered by Android and Gemini, developed with partners like Samsung and fashion brands. These glasses preview how Gemini’s multimodal understanding—especially from the Gemini Omni model—could anchor everyday, heads-up computing, blending AI assistance, search, and real-time context into a single wearable experience.
AI Infrastructure, Science Tools, and What Comes Next
Underpinning the Gemini announcements is a significant investment in AI infrastructure and scientific applications. Ahead of I/O, Google revealed a major cloud venture with Blackstone focused on expanding access to its custom Tensor Processing Units (TPUs). Backed by an initial equity commitment of USD 5 billion (approx. RM23.5 billion), the initiative aims to deliver 500 megawatts of data centre capacity starting in 2027, with Google supplying both hardware and software to the new company. Onstage, Demis Hassabis highlighted how Gemini-powered tools are already being used to build digital twins of the Earth for climate and weather simulations, and to accelerate drug discovery pipelines. Taken together—the TPU expansion, the scientific focus, and the new Gemini 3.5 Flash and Omni capabilities—Google I/O framed Gemini not just as a product family, but as a long-term platform for agentic AI, scientific research, and next-generation, multimodal user experiences.