Google launches Gemini 2.5, its most advanced AI model yet

The new version brings significant advancements in reasoning, programming, and multimodal understanding

Google announced this week the Gemini 2.5, its most intelligent AI model yet.

The new version, especially the Gemini 2.5 Pro Experimental, has achieved top performance in several benchmarks and demonstrated notable advancements in logical reasoning and code generation.

In today’s article, we’ll learn more about Gemini 2.5 and even create a Tetris game with the help of the new model.

AI that “thinks” before responding

Unlike previous models, Gemini 2.5 was designed to process information in a more structured and contextualized manner, ensuring more accurate responses aligned with the problem presented.

In the field of AI, the concept of "reasoning" goes beyond simple predictions or classifications. It involves contextual analysis, logical inference, and more sophisticated decision-making. Google has been refining this capability over the years with techniques such as reinforcement learning and chain prompting.

Now, Gemini 2.5 takes this concept to a new level, combining a more robust base model with a refined post-training process. This advancement will enable the creation of even smarter agents that are more sensitive to context.

The power of Gemini 2.5 Pro

Source: Chatbot Arena.

The Gemini 2.5 Pro Experimental is the most advanced version of the line and already leads the LMArena ranking, which evaluates AI models based on human preferences. Additionally, it has achieved superior performance in benchmarks for mathematics, science, and programming, solidifying itself as one of the most powerful AIs on the market.

Gemini 2.5 Pro sets a new standard in tests that require advanced reasoning, such as GPQA and AIME 2025. Without using inference techniques that increase cost, such as "majority voting," it excels in complex tasks involving mathematics and science.

Moreover, the model achieved 18.8% accuracy in Humanity’s Last Exam, one of the most rigorous tests of knowledge and human reasoning, created by experts from various fields.

Multimodal and Expanded Context

Gemini 2.5 retains the essence of the Gemini line, with native support for text, audio, images, video, and code. It arrives on the market with a context window of 1 million tokens, with plans for expansion to 2 million tokens soon.

This enables the model to process and understand large volumes of data, making it an even more powerful tool for developers and researchers.

More Efficient and Advanced Programming

Another significant leap is in the field of software development. Gemini 2.5 surpasses its previous version in code creation and editing, being capable of developing web applications, transforming legacy systems, and enhancing existing code with greater precision.

In the SWE-Bench Verified benchmark, a reference for evaluating autonomous code agents, Gemini 2.5 Pro achieved 63.8% accuracy in a customized configuration.

One of the highlights is the model’s ability to create interactive games from simple commands, generating executable code with minimal inputs, as shown in the model's presentation video.

And it’s this ability to create games in just seconds that we’ll explore now!

Creating a Tetris Game with Gemini 2.5

To explore the model’s potential, there’s nothing more fun than creating an interactive game! I’ve chosen the Tetris game, a classic, well-known, and highly enjoyable game.

First, let's access Google AI Studio.

When entering the platform, make sure the Gemini 2.5 Pro Experimental model is selected, as shown in the image above.

Next, send the command to the model and click on "Run."

Make me a captivating Tetris game. Key instructions on the screen. p5js scene, no HTML. I like pixelated blocks and interesting backgrounds.

In a few minutes, the model will generate a JavaScript code containing all the programming logic for the Tetris game, along with instructions on how to run the created game.

Copy the generated source code and paste it into the online editor p5.js so we can test our game.

p5.js is a JavaScript library geared toward generative art, graphics, and interactivity, based on the Processing language. It simplifies the creation of animations, games, and interactive visualizations directly in the browser.

Next, click on the "Play" button to run the created game.

The result? A super sleek and fully functional Tetris game!

It computes the scores correctly and even shows the shadow of where the piece will land! Impressive, right?

You can create as many games as you like, customizing them according to your preferences. Just specify it in the prompt.

I even recreated the dinosaur game that Google showcased in the Gemini 2.5 launch video. And guess what? It turned out perfect!

Availability and Next Steps

The release of Gemini 2.5 marks a significant advancement in artificial intelligence, opening up new possibilities for more sophisticated and efficient applications.

As demonstrated, in just a few minutes, it's possible to create complete applications – like the Tetris game we developed here – showcasing the power and versatility of this new model.

So, what do you think of the game we created? Share your thoughts! 🚀