Google doubles down on AI with Gemini, Veo and Imagen at I/O 2025

At its annual I/O developer conference in Mountain View, Google laid out its most ambitious AI roadmap yet — one that pivots on a universal AI assistant powered by upgraded Gemini 2.5 models and extends into creative tools like Veo 3 for video and Imagen 4 for image generation. Google CEO Sundar Pichai framed the moment succinctly: AI is no longer just emerging tech — it’s everywhere, evolving rapidly, and reshaping how people interact with digital tools. “We’re in a new phase of the AI platform shift,” he declared, highlighting decades of research now translating into real-world impact.

Gemini 2.5 Pro adds ‘Deep Think’ for advanced reasoning, while Veo 3 and Imagen 4 deliver powerful upgrades for video and image generation

The centrepiece is Gemini 2.5, now split into two distinct models — the powerful Gemini 2.5 Pro, and the leaner Gemini 2.5 Flash. The Pro version introduces a new ‘Deep Think’ mode for enhanced reasoning in maths and coding, while Flash sees improvements in speed, multimodality and contextual understanding. Flash is now available in experimental mode via Google AI Studio and the Gemini app, with Pro’s wider release pending further safety evaluations.

These models anchor Google’s vision for a universal AI agent, an assistant capable of reasoning, planning and performing tasks across apps and devices. DeepMind CEO Demis Hassabis described it as an AI that can “understand context and take action on your behalf”. It’s a culmination of internal efforts under Project Mariner and Project Astra, aiming to enable agents to handle up to ten tasks in parallel — from bookings to purchases and research.

Veo 3 and Imagen 4: Next-Gen generative media tools

Google also debuted its next-generation creative AI models. Veo 3 adds the ability to generate videos with background audio, realistic physics, and lip-synced dialogue. Users can now tweak elements like zoom and camera movement with more precision. Imagen 4, on the other hand, impresses with hyper-detailed rendering — capturing everything from animal fur to intricate fabrics in photo realistic or abstract styles.

It supports up to 2K resolution and delivers dramatically improved spelling and typography, making it ideal for posters, comics and visual content. Both models are being integrated into Google's wider ecosystem — with Imagen 4 available in Workspace apps like Slides and Docs, while Veo 3 is currently exclusive to Ultra subscribers in the US via the Gemini app and Google’s new AI filmmaking tool, Flow.