Google AI Updates: October 28, 2026

1. Google DeepMind Announces Genie 4 World Model With Native Audio and Gemini-Guided Simulation

Google. Google DeepMind announced Genie 4, a real-time interactive world model that generates audio natively alongside the world, including ambient sound and interactive dialogue from characters. The model builds on the Gemini family: Gemini Omni grounds appearance and behavior, and Gemini 3.8 Flash runs in the loop to break goals into steps, review incoming frames, and steer the scene, while users can change weather, camera, or add people and objects with text prompts mid-simulation. DeepMind also showed Gemini 4 Argon planning tasks that Genie 4 renders, such as simulating Apptronik’s Apollo humanoid playing checkers. Genie 4 is in research preview with trusted testers and is coming to Google AI Ultra subscribers in Project Genie and early access partners in the Gemini API and Google AI Studio; output carries SynthID watermarks, and DeepMind notes that object permanence and consistency still drift over longer sessions. Source

2. A2UI Adds Swappable Inference Formats and the Faster “A2UI Express” Default

Google. Google’s A2UI team released swappable inference formats for the A2UI generative UI protocol, letting agents generate a compact intermediate format that the SDK compiles into standard A2UI JSON messages instead of emitting the full JSON tree. The new default, A2UI Express, uses a Python-like one-statement-per-line syntax and, according to Google, needs up to 3x fewer tokens and cuts latency by up to 3x compared with direct JSON; Google reports it holds up even with model thinking turned off on a 51-task eval using gemini-3.5-flash. The post also shows pairing A2UI with decision models such as TypeSafe’s Jev for predefined layouts, with up to 80x lower latency. Express ships in the Python agent SDK (a2ui-agent-sdk 0.9.0), the TypeScript agent SDK (@a2ui/agent), and the Dart SDK. Source