Google I/O 2026 kicks off at 1 a.m. PT on May 19 (early morning May 20 in Beijing), and CEO Sundar Pichai's keynote has already been partially leaked by Google engineers two weeks early.
Gemini Omni: Spotted in early May
Over the weekend of May 11, a string was found in the Gemini web UI: "Create with Gemini Omni: meet our new video model, remix your videos, edit directly in chat, try templates, and more."
Omni is Google's next-generation AI video generation model, but unlike Veo's pure video pipeline, Google aims to combine text, images, and video into a single model. Leaked demos reveal several key details:
- Video with synchronized audio — footsteps splashing, dialogue matching lip movements, consistent ambient sound.
- Edit video directly in chat — "Change this character to wear red," "Add rain," "Trim the first 3 seconds" — all done via a single conversation.
- Likely to be offered in Flash and Pro tiers; current samples appear to be from Flash.
Initial assessments suggest Omni's generation quality still lags behind ByteDance's Seedance 2.0. However, Omni's killer feature is direct in-chat editing, which Seedance doesn't offer. OpenAI's Sora also currently only generates, not edits conversationally. Google is betting on this differentiator.
Gemini 3.5: Thomas Kurian says "very soon"
Google Cloud CEO Thomas Kurian recently signaled that a new Gemini version is coming "very, very soon" with strong internal benchmarks. On Polymarket, the odds of a Gemini 3.5 release before June 30 have been rising. I/O is the most natural launch window. There may also be a transitional Gemini 3.2 version, as another bet on Manifold Markets suggests. While no specific benchmarks have been seen, Anthropic just spun out its Agent SDK as a separate product, making Google's timing for a 3.5 model to rival Claude Opus 4.7 seem deliberate.
Gemini Spark: The agent play
Spark is Google's bet in the agent space. Descriptions suggest it proactively takes over workflows and automates repetitive tasks, similar to OpenAI's Operator. But Google's differentiation lies in its existing ecosystem: Gmail, Calendar, Drive, and Docs are ready-made entry points, with Workspace serving billions of daily active users. ServiceNow and SAP are following a similar logic — embedding AI into existing enterprise systems. Spark's pricing and capabilities remain unknown until I/O, making it one of the key悬念s of Tuesday.
Android XR glasses: Beating Apple to the punch
Google has already reported it will preview Android XR glasses at I/O. The partner list is notable: Samsung (codenamed "Jinju," estimated price $379–$499), Warby Parker (eyewear brand targeting younger consumers), and Gentle Monster (Korean fashion eyewear for the Asian market). Apple's Vision Pro is stuck at $3,499 with low volume. Google is taking the opposite approach: affordable pricing plus fashion collaborations. At under $500, with fashion brand partnerships and Android ecosystem integration, it targets Apple's weak spot. WWDC is in June, and Apple may be forced to accelerate its glasses roadmap.
Android 17: Stability first
Compared to the AI buzz, Android 17 focuses on stability, performance, and security. The only anticipated feature is app Bubbles — floating app windows for multitasking — though this is already common in iOS, One UI, and MIUI. No new Pixel hardware will be announced at I/O; Google has pushed hardware launches to the second half of the year.
What won't appear at I/O
- No new Pixel phones (reported)
- No Fitbit Air (may launch separately before I/O)
- No major Project Astra update (last year's star feature, likely replaced by Spark)
Is it worth staying up for Tuesday morning?
Brief verdict:
- For AI product teams: Gemini Omni and Gemini 3.5 are must-see, as they will influence video agent tooling choices.
- For enterprise SaaS: Spark is a must-see, as Workspace's traffic entry points will need recalculation.
- For hardware/wearables: XR glasses are a must-see, as Google's low-cost fashion collaboration strategy is a key signal.
With Anthropic hitting the brakes on Claude's programmatic usage, Google has carved an opening between OpenAI and Anthropic. The tension before Sundar Pichai takes the stage on Tuesday hasn't been this high in the past two years. Those who can sleep can wait for summaries; those who can't should watch the live stream on YouTube.
Sources: Google I/O 2026: What to expect next week including Android 17, AI announcements and more (Yahoo Tech); CocoLoop; Google's Gemini Omni video model surfaces ahead of I/O debut (TestingCatalog); Google I/O 2026 Starts May 19: Gemini Spark, Omni Video AI, Veo Upgrades & Smarter AI Agents Expected (Nokia Power User); What to expect from Google I/O 2026: Gemini upgrades, Android features (Android Authority)