Gemini 3.5 Flash, Gemini Omni, and Gemini Spark headline I/O 2026

Google's I/O 2026 keynote was the most consequential model-launch event of the week. Gemini 3.5 Flash is positioned as Google's most capable coding and agentic model yet — four times faster than the prior frontier generation — and ships immediately into the Gemini app, AI Mode in Search, and the new Antigravity 2.0 developer platform. Logan Kilpatrick described Omni as the company's new "create-anything-from-any-input" model, starting with video and slotted into Flow and YouTube with API access coming soon.

Gemini Spark is the headline consumer product: an always-on personal agent that runs on 3.5 Flash and uses Antigravity to orchestrate sub-agents handling email triage, scheduling, research, and digital errands. Google paired the release with a sweeping subscription overhaul — three tiers from $7.99 to $99.99/month — and a shift away from daily prompt limits toward consumption-based compute pricing, matching OpenAI's and Anthropic's enterprise economics.

DeepMind also rolled out "Co-Scientist," a multi-agent system for accelerating scientific hypothesis generation (exposed via a new "Hypothesis Generation" experimental tool), and previewed Project Genie integrated with Street View for immersive world simulations for robotics and gaming. The 900M-MAU figure — more than double year-over-year — and AI Overviews hitting 2.5B users mark a definitive turn: Gemini is now distribution-led at consumer scale, not playing catch-up.

Developers' biggest gripe was the delay of Gemini 3.5 Pro ("Cappuccino") to next month, and Ethan Mollick noted that Google's product surface is diverging — Studio, the Gemini app, Antigravity, and consumer Gemini are becoming distinct experiences, while Anthropic and OpenAI converge on unified Code/Cowork stacks. The competitive takeaway: Google is now matching frontier model quality AND owning the agent runtime AND the search surface — a stack OpenAI lacks and Anthropic isn't building.