Google I/O 2026 unveils Gemini 3.5 Flash, video 'world model' Gemini Omni, and autonomous agent Gemini Spark

At Google I/O 2026 (keynote May 19-20), Google announced a new tier of Gemini models centered on speed, multimodality and autonomy. Gemini 3.5 Flash, billed as combining frontier intelligence with agentic action, was reported to output tokens roughly four times faster than other frontier models at its tier and to outperform Gemini 3.1 Pro on agentic and coding suites, with Google citing Terminal-Bench 2.1 at 76.2%, GDPval-AA at 1656 Elo, and MCP Atlas at 83.6%; it began rolling out immediately across the Gemini app, Search, the Gemini API and Antigravity 2.0. Gemini Omni was introduced as a 'world model' that can generate output across formats from any input, starting with video, and described as combining an intuitive understanding of physics (gravity, kinetic energy, fluid dynamics) with Gemini's broader knowledge; generated video carries SynthID watermarking.

The Omni Flash variant began rolling out to Google AI Plus, Pro and Ultra subscribers via the Gemini app and Google Flow, and appeared in YouTube Shorts Remix and the YouTube Create app at no cost for users 18+. Google also previewed Gemini Spark, an autonomous personal agent built on Gemini 3.5 and Google Antigravity that 'works in the background' on a phone or laptop and acts under user direction, opening first to trusted testers and US Google AI Ultra subscribers. Google said Gemini 3.5 Pro was being used internally and in testing, with broader availability expected the following month (June 2026).

The slate underscores Google's push from chat assistant toward agents and generative 'world models,' and intensifies the speed/cost competition against OpenAI and Anthropic at the Flash tier. Benchmark figures are Google-reported.