Google I/O 2026 Opens at Shoreline as Sundar Pichai Unveils Gemini 4 Omni Multimodal Model and Previews Android XR Glasses

Level 2 — Elementary

Google held its biggest yearly event, called Google I/O, on Tuesday, May 19, 2026. The CEO of Google, Sundar Pichai, gave the main speech at the Shoreline Amphitheatre in Mountain View, California, starting at 10 a.m. local time.

The biggest announcement was a new artificial intelligence model called Gemini 4 Omni. Earlier Gemini models could understand pictures or sound, but Gemini 4 Omni works with text, pictures, video, audio and computer code all at the same time. It can remember a very long conversation — up to two million words — without forgetting the beginning.

Google also showed a small pair of computer glasses for the first time. They run a smaller version of Gemini directly inside the glasses, so users can ask questions and the glasses can describe what they are looking at. The glasses are made by Samsung with chips from Qualcomm.

Other announcements included Aluminium OS, a new laptop operating system that uses ARM-style chips like a smartphone, and a feature called Gemini Intelligence that will arrive on Pixel and Samsung phones during the summer.

annual: happening once every year

CEO: chief executive officer; the most senior leader of a company

artificial intelligence: computer systems that can do tasks usually done by people, such as learning, deciding or answering questions

model: in AI, a trained computer program that can answer questions or do tasks

multimodal: able to work with several different types of information, such as text, sound and images

operating system: the main software that makes a computer or phone work

ARM: a family of computer-chip designs widely used in smartphones because they save energy

preview: an early showing of something before it is fully released

Level 3 — Intermediate

Google opened its annual I/O developer conference at the Shoreline Amphitheatre in Mountain View, California on Tuesday, May 19, 2026, with chief executive Sundar Pichai delivering a ninety-minute keynote anchored by the formal release of Gemini 4 Omni, the company's first end-to-end multimodal foundation model. Unlike the Gemini 3 family, which routed image, audio and video through separate encoder pathways before fusing into a shared text decoder, Gemini 4 Omni processes every input modality in a single unified transformer with a native two-million-token context window — long enough to ingest a feature-length screenplay, a full codebase or several hours of high-bitrate video in one prompt.

Pichai also previewed the first Android XR glasses reference design — a 38-gram titanium-frame head-mounted display jointly engineered with Samsung Electronics and powered by Qualcomm's new XR2 Plus Gen 4 chip — running a distilled on-device variant called Gemini Nano XR. Demonstration video showed real-time translation overlaid on a conversational partner, turn-by-turn navigation inside Tokyo's Shinjuku Station, and image-based plant identification, all delivered without cloud round-trips. Samsung confirmed a launch partner programme with retail availability through Verizon and AT&T from late summer 2026.

On the consumer software side, the keynote announced Gemini Intelligence as a system-level agentic layer for Pixel 10 and Samsung Galaxy S26 series devices, rolling out via OTA in July; Aluminium OS, a new ARM-native laptop platform that supersedes ChromeOS for higher-end notebooks; and a series of updates to Workspace including Gemini in Sheets for SQL-style natural-language analysis and a complete rewrite of Google Meet's noise-cancellation stack using on-device generative models.

For developers, the technical keynote that immediately followed unveiled new pricing for Gemini 4 Omni via the Gemini API at $1.25 per million input tokens and $5.00 per million output tokens, undercutting OpenAI's GPT-5.5 by roughly 35% on input and 22% on output; the addition of native function-calling for arbitrary HTTP webhooks; and a tighter Vertex AI integration that places fine-tuned Gemini 4 models on dedicated H200 and TPU v6 hardware tiers.

developer conference: an event where a technology company shares new products and tools with software developers

keynote: the main speech or presentation at a conference, usually given by a senior executive

foundation model: a large machine-learning model trained on broad data that can be adapted to many downstream tasks

transformer: the neural-network architecture that underpins most modern large language models, based on attention mechanisms

context window: the maximum amount of text or tokens a language model can consider at once when generating a response

on-device inference: running an AI model directly on the user's hardware rather than in a cloud data centre, improving privacy and latency

OTA: over-the-air; a software update delivered to a device through its wireless internet connection

Vertex AI: Google Cloud's managed machine-learning platform for training, deploying and monitoring AI models

Level 4 — Advanced

Google opened its annual I/O developer conference at the Shoreline Amphitheatre in Mountain View, California at 10:00 a.m. Pacific Daylight Time on Tuesday, May 19, 2026, with chief executive Sundar Pichai delivering a ninety-three-minute opening keynote anchored by the formal general-availability release of Gemini 4 Omni, the company's first end-to-end natively-multimodal foundation model. Unlike the Gemini 2.5 and 3.x families — which routed image, audio and video through separately-trained encoder pathways before late-fusing into a shared text decoder — Gemini 4 Omni processes every input modality through a single unified mixture-of-experts transformer with byte-level patch embedding for non-text inputs, a native two-million-token context window, and SoftMoE routing across 184 expert blocks.

Pichai also previewed the first Android XR glasses reference design — a 38-gram titanium-frame head-mounted display jointly engineered with Samsung Electronics' Mobile eXperience division and powered by Qualcomm's new XR2 Plus Gen 4 SoC fabricated on TSMC N3P — running a distilled four-billion-parameter on-device variant marketed as Gemini Nano XR. Demonstration video showed real-time bidirectional Japanese-English speech translation overlaid as captions on a conversational partner, sub-200-millisecond turn-by-turn navigation inside Tokyo's Shinjuku Station Yamanote Line concourse, and image-grounded plant identification against the iNaturalist taxonomy, with all inference happening without cloud round-trips and a 30 mW peak compute envelope. Samsung confirmed a launch partner programme with retail availability through Verizon and AT&T from late August 2026 at a $1,299 starting price.

On the consumer-software side, the keynote announced Gemini Intelligence as a system-level agentic execution layer for the Pixel 10 series and Samsung Galaxy S26/S26+/S26 Ultra, rolling out via OTA in July 2026 under the codename 'Astra'; Aluminium OS, a new ARM-native laptop platform jointly developed with Qualcomm, MediaTek and Samsung System LSI that supersedes ChromeOS Flex on higher-end notebooks and ships with a native Linux subsystem and an Android-app compatibility layer; and a series of updates to Workspace including Gemini in Sheets for SQL-style natural-language analysis, a Gemini-in-Docs collaborative-rewrite mode, and a complete rewrite of Google Meet's noise-cancellation stack using an on-device generative source-separation model trained on 480,000 hours of multilingual speech.

For developers, the technical keynote immediately following Pichai's opener unveiled Gemini 4 Omni pricing via the Gemini API at $1.25 per million input tokens and $5.00 per million output tokens (with a separate, lower-cost video-token rate of $0.85/million), undercutting OpenAI's GPT-5.5 by roughly 35% on input and 22% on output and Anthropic's Claude Opus 4.6 by roughly 50% across the board; native function-calling for arbitrary HTTPS webhooks with parallel-tool-invocation support; and a tighter Vertex AI integration that places fine-tuned Gemini 4 Omni checkpoints on dedicated NVIDIA H200 and TPU v6 'Trillium-Plus' hardware tiers with provisioned-throughput discounts of up to 42% under three-year reserved commitments. Wall Street reacted favourably: Alphabet's Class A shares closed up 3.8% on Nasdaq at $217.40, taking the company's market capitalisation past 2.7 trillion dollars.

mixture-of-experts transformer: a neural-network architecture in which a learned router selects a subset of specialised sub-networks (experts) to process each input token, lowering inference cost at large scale

byte-level patch embedding: a technique for representing non-textual inputs to a transformer by partitioning raw bytes into fixed-size patches that are linearly projected into the model's token-embedding space

SoftMoE routing: a soft-routing variant of mixture-of-experts that assigns each input to a weighted blend of experts rather than discrete top-k selections, improving training stability

system on chip: an integrated circuit that combines processor cores, memory controllers, graphics, modem and AI accelerators on a single die, typically used in mobile and wearable devices

TSMC N3P: Taiwan Semiconductor Manufacturing Company's 3-nanometre-class second-generation process node, offering ~5% performance improvement at the same power versus N3E

compute envelope: the maximum power and thermal budget that hardware can dissipate while operating, typically expressed in watts

provisioned-throughput discount: a cloud-pricing instrument that offers reduced per-token rates in exchange for a commitment to sustained inference capacity over a fixed term

generative source separation: a class of audio-processing model that learns to isolate specific sound sources (e.g. a target speaker) from a noisy mixture by generating the clean signal directly rather than masking the input

Level 1 — Absolute Beginner

Level 2 — Elementary

Level 3 — Intermediate

Level 4 — Advanced

Google I/O 2026 Opens at Shoreline as Sundar Pichai Unveils Gemini 4 Omni Multimodal Model and Previews Android XR Glasses

Multiple Choice

True or False

Fill in the Blank