Google I/O: AI subscription for $ 250 and an agent Gemini

Intelligent, “agent” and personalized – these are the new attributes for AI, which Google’s CEO Sundar Pichai said at the start of the Google I/O developer conference. Behind it is a general trend: the AI ​​should be able to answer even better, more correctly and detailed questions, but also do tasks themselves and always act in the interests of the user. And at some point, as Demis Hassabis, CEO from Google Deepmind, we then have a world model that can do everything – like a real brain.

Google’s most powerful models Gemini 2.5 Flash and Pro already lay the foundation for this new AI world, according to Hassabis. Google wants to have improved it further and make more people accessible. Gemini 2.5 Flash moves into the Gemini app and is available for developers via Google Ai Studio and Vertex AI – 2.5 Pro should follow soon. Google had already made a preliminary version available for developers. According to Google, Gemini 2.5 Pro is a leader in the WebDev Arena with an ELO of 1420 and in all LMarena categories-AI models compete against each other in such an arena.

Gemini can think, speak and control the browser

Deep Think, a so -called Reasoning model that is in Gemini 2.5, is also new. Reasoning means that the model should not only be able to reproduce content, but also link them logically. Both models can soon generate native audio. The key, an accent and the style of a speaker can be determined. Whisper, speak dramatically, everything possible.

Gemini 2.5 flash and Pro are also suitable as agents that can control the browser. Google has been working on this for a long time under the name Project Mariner. This now moves into the Gemini API and Vertex Ai. Google also supports the Model Context Protocol (MCP) that Anthropic has developed. Thanks to the MCP, Gemini can collect information and create lists, such as where there is a laundry. The protocol makes websites “readable” for agents.

$ 250 for a VIP pass

Project Mariner is already available for customers of Google’s new AI Ultra subscription in the USA. Google calls this a “VIP pass”-and also makes it pay: it costs $ 250 a month. Including are the highest usage limits for all AI services, for example, they also get access to Deep Think, Flow and whisk, for example, with which you can transform images into short animations. The subscription also includes 30 TB storage capacities for photos and other documents. AI Premium is renamed Ai Pro – you can keep it and get more access here, for example on flow.

The search is expanded to include AI Mode in the USA. So far, it has already been available in a test environment of the Google Labs. The Reasoning capabilities are also behind the AI ​​mode. This should be able to answer more complex questions in the search. To this end, Google also uses the web its own knowledge graph, i.e. the knowledge that Google has collected in a gigantic database. The shopping results are also in this database. So it happens that Google Shopping gets AI functions through AI Mode. This includes the possibility to try out clothing virtually, Google had tested something like this earlier, and there is also an agent checkout.

Images 4 and VEO 3 are Google’s new and improved video and image generators. VEO 3 can generate native audio in the future, for example text and ambient noise. Flow should have longer films created, for example by getting better control over characters and styles. Google seems to want to prove this with a video at the start of the I/O. A zoo whirls up a classic wild-west city. Parrots fly around, a T-Rex from clamping modules roars, it dusts. But the balloon -like letters that welcome the I/O shine clean cleanly in the blue sky.

Canvas, Google’s Ki-Skribbletool, Deep Research, the chatbot for scientific tasks, get updates, as well as Gmail and Google Meet-this includes emails that answer Gemini as you write yourself and synchronous translations in Meet. At first there are only a few languages.