Google I/O '25 KeynoteMay 20, 2025 · 1h 55m · Sundar PichaiGoogle announced advances in AI agents, multimodal models and personalized Gemini experiences, citing over 480 trillion monthly tokens processed and 400 million Gemini app users.
Key themes: Models, Assistants, Agents
- 1Google now processes over 480 trillion tokens per month across its products
- 2The Gemini app passed 400 million monthly active users
- 3AI Mode in Search rolled out to all US users with deeper reasoning via Gemini 2.5
- 4Veo 3 generates video with native, synchronized audio
- 5Imagen 4 became the latest image generation model available in the Gemini app
- 6Gemini 2.5 Pro gained an experimental Deep Think reasoning mode
- 7Agent Mode in the Gemini app can handle multi-step tasks like apartment hunting
- 8Google Beam, an AI 3D video communication platform, is coming later in 2025 with HP as first hardware partner
| Announcement | Status |
|---|---|
| Google BeamAI-powered 3D video communications platform; first devices via HP | Coming |
| Gemini LiveCamera and screen-sharing on Android, rolling out to iOS | GA |
| Agent Mode (Gemini app)Experimental; automates tasks like apartment hunting | Coming |
| AI Mode (Search)Available now in the US with Gemini 2.5 integration | GA |
| Gemini 2.5 Pro Deep ThinkExperimental reasoning mode | Preview |
| Veo 3Video generation with native audio, in the Gemini app | GA |
| Imagen 4Latest image generation model | GA |
| FlowAI filmmaking tool for cinematic content | GA |
Companies on stage
Sources: official keynote videos, their captions and the hosts' recaps, summarized by AI; watch the source for exact wording. Logos via logo.dev; trademarks belong to their owners.