Google I/O 2026 Opens Tomorrow: New Gemini Model, Omni Video Generation, and Project Astra on Stage
Google I/O 2026 keynote starts May 19 at Shoreline Amphitheatre. Confirmed on the slate: a new Gemini version, a video-generation model called Gemini Omni, and broad 'Gemini Intelligence' integrations across Android and hardware. Leaked benchmarks suggest the new model lands below Claude Mythos and GPT-5.5.
Google I/O 2026 opens tomorrow, May 19, at 10 a.m. PT at Shoreline Amphitheatre in Mountain View. The developer keynote follows at 1 p.m. PT. This is the show where Google typically fires its full competitive arsenal — and after a week of pre-announcements at “The Android Show” on May 12, the remaining ammunition appears to center on Gemini.
What’s confirmed
A new Gemini version is on the keynote schedule. Analysts tracking Google’s internal versioning consider 3.2 or 3.5 the most likely designations, though a jump to Gemini 4.0 branding has not been ruled out. Separate from model capability, the “Gemini Intelligence” system — announced at The Android Show — is receiving its formal launch: proactive AI running persistently across Android phones, Wear OS watches, Android Auto, and the XR glasses hardware revealed last week.
“Gemini Omni” is the name attached to a video generation and editing capability expected to debut on stage. Unlike standalone video generation products, Omni is described as operating directly inside the Gemini chat interface — users can generate, trim, and refine video clips within a conversation thread without switching applications.
Project Astra, Google DeepMind’s real-time multimodal AI assistant, is also on the agenda. Astra demos at I/O 2025 drew significant attention; this year’s iteration is expected to show persistent memory and tighter integration with Google services.
The benchmark context
Leaked evaluation results circulating among AI researchers this week put the incoming Gemini model below both Anthropic’s Claude Mythos — which benchmarked at 94.6% on GPQA Diamond in partner previews — and OpenAI’s GPT-5.5. If accurate, this matters: Google’s I/O models have generally matched or beaten rivals at announcement. Falling short of two competitors simultaneously, with publicly confirmed benchmarks, would be a first.
Google has not commented on the leaks. The company’s position has consistently been that benchmarks tell a partial story and that real-world performance in Search, Workspace, and Android integrations is the relevant measure. That argument is harder to make when enterprise developers choosing API providers use benchmark scores as a primary filter.
The hardware angle
Googlebook laptops — Android-based AI-first machines announced at The Android Show with Acer, ASUS, Dell, HP, and Lenovo as hardware partners — are expected to receive a formal shipping timeline tomorrow. The devices run Aluminium OS (a ChromeOS evolution), with Gemini Intelligence as the primary interface. Android 17 and Android XR are likely to receive more detail than the preview shown last week.
What developers are watching
The Gemini API’s context window, pricing tiers for Gemini 2.5 Pro versus the incoming model, and any changes to the Vertex AI integration are the practical stakes for developers. Google’s code execution capabilities in Gemini, currently competitive with Claude Code at specific tasks, may get a significant upgrade on stage.
The full keynote streams on YouTube starting 10 a.m. PT May 19.