New functionalities

Artificial Intelligence, Gemini 2.0 and 2.5: many new functions, even for free. Here's what and why to use them

DeepMind has developed the Gemini family of models to meet a wide range of needs. These range from everyday use for various doubts to professional and scientific applications

12 April 2025

4' min read

Google surprised many in March by also offering free users many advanced functions that it previously jealously guarded, exclusive to those who paid. Particularly worth trying is the Deep Research function, which gives detailed reports with lots of sources, and, for those with scientific purposes, the Gemini 2.5 Pro (now experimental). It must be said that those who pay (in the Advanced version at EUR 21.99 per month, which also includes 2 Terabytes of cloud space) still have more: the ability to make more requests (technically 'tokens') and to enter more information to be analysed ('larger context').

In short, Gemini is advancing very fast, now at version 2.0 (2.5 for the Pro function).

What is Gemini and what it includes

DeepMind has developed the Gemini family of models to meet a wide range of needs. These range from everyday use for various doubts to professional and scientific applications. Gemini includes various models, successors to Google's own LaMDA and PaLM 2, which the user can choose from, depending on various needs. They are 'multimodal' in the native sense, i.e. designed to simultaneously process text, images, audio, video and code. As mentioned, since March, all models and functions are also available to free users, but the Advanced subscription offers enhanced resources and functionality.

For tasks requiring immediate interaction, Google has made available Gemini Flash, the template designed to ensure quick and responsive responses. This model is ideal for tasks such as brainstorming or copywriting, where speed is of the essence and articulated reasoning is not required. Gemini Flash offers high performance with low resource consumption. It is therefore good for everyday use. With Flash, we can also select the Canvas function, which serves as an interactive space for creating and editing text and code content. This platform allows users to collaborate with artificial intelligence in real time, facilitating document drafting, programming and creative projects.

When the task requires more in-depth thinking, we can try Gemini Flash Thinking, Google's lighter 'reasoning' model, now offered in an 'experimental' version. This variant of Flash incorporates an internal 'chain-of-thought prompting' process, which allows the model to break down the problem into several logical steps before providing the final answer. In this way, the model is able to handle complex, multi-step problems. It should therefore provide greater precision without compromising too much on speed.

A further advanced feature offered by the Gemini family is Deep Research. Although not a stand-alone model, this user-selectable function transforms a complex query into a multi-step research plan. The system autonomously scans the web, gathering and synthesising information from numerous sources to produce a detailed report. The user can export the report to Google Docs or turn it into an 'Audio Overview', i.e. an artificial intelligence-generated podcast. The free version has some limitations, such as a reduced contextual window (about 1 million tokens) and restrictions on the number of requests per minute, which can slow down operations for particularly complex tasks.

ChatGpt has a similar function, which is also called Deep Research (big tech does not win the fantasy prize), but it is only for 'Plus' users, for a fee.

The wonders that big tech includes in its model announcements - each one promoted as 'more perfect' than the last - do not deceive us, however. No Deep Research exempts us from the obligation to verify what the AI produces, which continues to make mistakes, even trivial ones. For instance, we had asked Deep Research to describe the current functionalities of Gemini and, after a few minutes of lucubration, he continued to talk to us about Flash 1.5; even after we had told him to check, pointing out that by now there was 2.0 for all users, he persisted in pointing to it as the basic model for those who do not pay. Gemini had, in short, failed to realise that newer sources - including Google's own site - were replacing older ones.

However, Google's pinnacle of innovation in AI is now called Gemini 2.5 Pro, the model defined by the company as the most intelligent currently available. This model uses advanced reasoning techniques to analyse and solve complex problems. Its ability to handle multimodal input - ranging from text to images, audio to code - in a single session makes it suitable for business applications designed to optimise and automate complex, large-scale production processes.

But also for tasks requiring high precision, such as advanced coding, mathematical analysis or the management of articulated documents. The context window of Gemini 2.5 Pro, initially set at 1 million tokens, can be extended to 2 million in the Advanced version.

In addition, Advanced users have priority access to premium features, such as integration with other Google applications (e.g., Gmail and Docs), additional storage space via Google One, and other experimental options that make the experience much more complete and suitable for professional environments.

Since March, Gems have been available to everyone: they allow you to customise Gemini to create your own personal AI expert on any subject. We can start with one of the ready-made gems or quickly create our own customised gems, such as a translator, meal planner or maths coach. Just go to the "Gem Manager" in the left-hand menu on the desktop, write down the instructions, give it a name and then chat with it whenever you want. When creating a customised Gem you can also upload files so that it can refer to more useful information.

Brand connect

I prossimi eventi

Tutti gli eventi

Notizie e approfondimenti sugli avvenimenti politici, economici e finanziari.

Technology