Google rolls out new Gemini 2.5 updates with Agent Mode, Deep Think, and learning tools
Computex 2025At I/O 2025, Google has dropped a major set of updates for its Gemini app. The company announced new capabilities like Gemini 2.5 Flash, a more advanced Deep Think mode, a ChatGPT-style Agent Mode, and support for quiz generation, audio overviews, and document-based research - all bundled into the same ecosystem.
The biggest upgrade right now is Gemini 2.5 Flash, which brings smarter reasoning and more efficiency. It runs 20-30% lighter on token usage and is now live in the Gemini app for all users. Developers and enterprise customers can also access an updated version through Google AI Studio and Vertex AI as of May 20. The fully fledged Gemini 2.5 Pro model is expected to launch in early June.
Meanwhile, Gemini 2.5 Deep Think is being pitched as a reasoning-focused upgrade, showing stronger performance across math (USAMO 2025), coding (LiveCodeBench v6), and multimodal benchmarks (MMMU). Google says this mode will undergo more safety testing before being rolled out to the public, but it’s already in testing for select Gemini API users.
If you’re a Gemini Ultra subscriber, you’ll soon get access to Agent Mode, powered by something called Project Mariner - where you tell Gemini your objective, and it pulls from live web searches, your Google apps, and external tools to get it done. This feature lives in a split view where chat runs on the left and a browser-like panel handles content on the right.
On the creative side, Gemini Canvas has also picked up a new "Create" menu, letting users generate full webpages, infographics, quizzes, or audio overviews directly from chat. You can also describe a custom app and have Gemini sketch a starting point. And for students or professionals, a Deep Research tool now lets you combine public data with private PDFs, Google Drive files, and market trends.
Other new touches include:
Source(s)
