Alphabet (GOOG) Google I/O '25 Keynote summary

Key announcements and product launches

• Introduced Gemini 2.5 Pro and Gemini Flash, setting new benchmarks in AI performance and efficiency.

• Launched Google Beam, an AI-first 3D video communications platform, and announced its partnership with HP for device rollout.

• Rolled out real-time speech translation in Google Meet and expanded Project Astra and Mariner capabilities for AI assistants.

• Debuted Agent Mode in the Gemini app, enabling multi-step actions and integration with services like Zillow.

• Announced Gemini SDK compatibility with Model Context Protocol and new developer tools for agent ecosystems.

AI model advancements and infrastructure

• Gemini 2.5 Pro leads industry benchmarks, with significant improvements in coding, reasoning, and multimodality.

• Ironwood TPU delivers 10x performance over previous generations, supporting rapid AI model deployment.

• Gemini models now process 480 trillion tokens monthly, a 50x increase year-over-year.

• Gemini 2.5 Flash offers 22% efficiency gains and introduces thinking budgets for cost and latency control.

• Deep Think mode in Gemini 2.5 Pro enhances reasoning and is being tested for safety before broad release.

User adoption and ecosystem growth

• Over 7 million developers have built with the Gemini API, with 5x growth since last year.

• Gemini app surpassed 400 million monthly active users; AI Overviews in Search now reach 1.5 billion users monthly.

• Project Mariner and agentic capabilities are being integrated into Chrome, Search, and Gemini app.

• New AI subscription plans launched: Google AI Pro and Google AI Ultra, offering higher limits and early access.