Google I/O 2026: AI Ultra Plan and Gemini's Transformation Into an AI Hub

At Google I/O 2026, alongside the previously announced major releases marking the "year of agents," a raft of additional new products and services were unveiled.

The most notable announcement is the new pricing plan, the **AI Ultra Plan**. At $100 per month, it offers five times the usage quota of the existing AI Pro Plan. Introduced alongside the release of the Antigravity 2.0 desktop app and CLI tools, the plan is positioned for heavy users.

The **Gemini app** has received a major update, aiming to transform from a mere chatbot into an "all-purpose AI hub." Positioned to compete with ChatGPT and Claude, it seeks to become the center of everyday AI usage.

The announcement of **audio glasses** also drew attention. Inspired by Meta's smart glasses, the device lets users control Google apps and services through Gemini via voice commands. The display-free approach—hence the name "audio glasses"—appears to be a strategy for balancing practicality with everyday wearability.

Google also declared its full-fledged entry into the **AI design tool** space, with accessibility-focused design aimed at a broad audience ranging from teachers to small business owners.

**Gmail** is getting expanded AI features as well, including voice-based conversational email search. Just ask Gemini to "find that email about that thing," and it can dig up details from long-buried messages.

Google also announced a new feature called **Information Agents**. It monitors topics in the background and proactively notifies you of updates and changes, aiming to automate personalized information gathering in a way that goes beyond the traditional concept of search.

In the shopping space, Google announced the **Universal Cart**, a vision for unifying the shopping experience across multiple devices and stores and tracking your shopping journey across the entire internet.

SynthID Becomes the Industry Standard for AI Watermarking

**SynthID**, the AI-generated content watermarking technology developed by Google, is starting to be adopted by major players including OpenAI and Nvidia. The technology, which embeds invisible watermarks in AI-generated text and images, is becoming critical infrastructure for identifying the provenance of AI content.

While standardization advances across the industry, open-source tools designed to strip watermarks have also surfaced on GitHub, giving rise to what looks like a technological cat-and-mouse game. The spread of SynthID is an important step toward AI transparency, but a fundamental solution will require sustained technical and institutional responses.

AI Allegations in Literary Prizes Become Routine

It has been reported that **three of the five** regional winners of the prestigious **Commonwealth Short Story Prize** are **suspected of using AI**. According to WIRED's reporting, this is far from an isolated case, signaling that AI allegations in literary prizes are becoming "the new normal."

As the quality of AI-generated writing improves, how to guarantee the authenticity of creative works has become a challenge for the entire cultural sector.

Google AI Edge Gallery Adds Support for Gemma 4 MTP

Google's on-device AI app, **AI Edge Gallery**, has been updated to v1.0.13 and v1.0.14. The biggest change is support for **Multi-Token Prediction (MTP) in Gemma 4**, which is expected to speed up local inference.

The updates also include TPU support for Pixel, experimental MCP (Model Context Protocol) support, new features, and chat history saving—further improving the local LLM experience on mobile.

New SOTA 1B Model "HRM-text" Emerges

A model claiming new SOTA (State of the Art) results in the 1B parameter class, **HRM-text**, is drawing attention from the community. While its benchmark scores are highly impressive, community reactions have been skeptical, with some calling it "too good to be true." Its actual performance will have to await independent verification, but there is no doubt that competition among small models is heating up.

Ocean Raises $28M to Double Down on AI Phishing Defense

**Ocean**, an agentic email security platform, has raised $28 million from Lightspeed Venture Partners. Its founder has been active as a hacker since their teenage years and was also involved in research on Israel's missile defense system, the Iron Dome.

As AI-powered phishing attacks grow increasingly sophisticated, an "AI vs. AI" security front—fighting AI with AI—is now opening up in earnest.

---