Demis Hassabis unveils Gemini Omni AI that simulates physical world
At Google I/O 2026, DeepMind's Demis Hassabis introduced Gemini Omni, an AI that can actually understand and simulate the physical world: think turning tricky science ideas into visuals you can see.
It blends brainpower from Google's Veo, Nano Banana, and Genie models, making it a big leap toward smarter AI.
Omni creates videos, Gemini 3.5 flash
Hassabis showed off Omni by prompting it to create a clay animation video explaining protein folding. Pretty cool for science fans.
The model also lets you edit videos with simple prompts, like transforming shapes or switching up styles.
Plus, Google rolled out Gemini 3.5 Flash: It's now the default in the Gemini app and runs four times faster than other frontier models. Sundar Pichai says it's got better coding skills and speeds up software development at Google through Antigravity 2.0.