Google launched Gemini 2.0, a “comprehensive” AI model that can replace humans

Google’s new AI model, Gemini 2.0, has been unveiled and is extremely versatile—generating text, sound, and images, and offering new multimodal capabilities that lay the foundation for the next big step in AI: agents that can literally replace the user in routine operations. The new model has also become noticeably more productive and energy efficient.

Image source: techspot.com

Like every other company in the AI ​​race, Google is feverishly integrating AI into everything it can get its hands on in an attempt to create commercially successful products. At the same time, it is necessary to configure the entire infrastructure so that expensive AI solutions do not ruin the company. Meanwhile, Amazon, Microsoft, Anthropic and OpenAI are pouring their own billions into virtually the same set of problems.

Gemini 2.0 comes approximately 10 months after the release of version 1.5. Google DeepMind CEO Demis Hassabis praises the new model, calling it “comprehensive,” although Gemini 2.0 is still in the “experimental preview” stage, in Google’s terminology. However, Hassabis is confident that the new model will provide a completely different level of capabilities, primarily in the field of agent AI.

Agent AI refers to AI bots that can fully perform actions on behalf of the user. For example, Project Astra from Google is a visual system that can recognize objects, help you navigate the world and find lost items. According to Hassabis, Astra’s capabilities have increased exponentially in Gemini 2.0.

Another example is Google’s Project Mariner, an experimental extension for the Chrome browser that can literally spoof the user while surfing the web. The Jules agent, in turn, helps developers find and fix bad code. There’s even an agent released that helps you play video games better. Hassabis cites it as an example of a truly multimodal AI model.

«We really see 2025 as the real beginning of the agent-based era,” said Hassabis, “Gemini 2.0 is the foundation of that.” He also noted the increased performance and energy efficiency of the new model, especially against the backdrop of a general slowdown in progress in the AI ​​industry.

Google’s plan for Gemini 2.0 is to use it absolutely everywhere. Google’s goal was to pack as many features into a single model as possible, rather than launch many separate, disparate products. “Multimodality, different kinds of outputs, functions—the goal is to incorporate all of this into the underlying Gemini model. We are trying to build as general a model as possible,” says Hassabis.

As the agency era begins, Hassabis says AI will need to solve both new and old problems. The old ones are eternal, they concern productivity, efficiency and cost of output. New ones are largely associated with security and privacy risks.

Gemini 2.0 is currently at an experimental stage, and only in a lightweight version of Gemini 2.0 Flash. The final version is scheduled for release early next year.