Descriptions:
Paige Bailey, who leads engineering for developer relations at Google DeepMind, delivers a fast-paced live presentation at the AI Dev 26 x SF conference cataloging everything Google has shipped in the past few months. In a single slide she covers Gemini 3.1 Flash Live (real-time conversational model), Gemini 3.1 Pro and Flash Lite, Nano Banana 2 (interleaved image-text generation), a new multimodal embeddings model that places video, audio, images, and text in a shared vector space, Liria 3 for music generation, Genie3 for world-model building, VO3.1 Light (realistic video with audio at a fraction of the original cost), and Gemini 3.1 text-to-speech with multi-language and multi-style support.
The second half of the talk is a live walkthrough of Google AI Studio, showing how to attach YouTube video clips as context, run sandboxed Python via code execution, use URL-based retrieval, and ground responses with Google Search — and how clicking “Get Code” instantly produces API-ready snippets for any demo. Bailey then spotlights Gemma 4, Google’s newly Apache 2-licensed open model family, which spans a 2-billion-parameter version small enough for mobile devices, a 4-billion version that fits on a laptop, and a 26-billion mixture-of-experts and 31-billion dense model that rank competitively in the chatbot arena.
A practical note for developers: the free tier grants 10,000-plus daily API requests to Gemma models — a detail Bailey says Google does not widely advertise. The session is a useful single-source reference for anyone mapping Google’s current AI portfolio.
📺 Source: DeepLearningAI · Published May 20, 2026
🏷️ Format: Keynote Launch
