New Version of Google’s Artificial Intelligence Doubles Performance on Benchmark, Strengthens Multi-Step Automation and Expands Creation of Dashboards and 3D Simulations
Google announced, in 2025, the Gemini 3.1 Pro, an update to the Gemini 3 series aimed at advanced logical reasoning and complex tasks.
According to official data released by the company itself, the model increased its performance on the ARC-AGI-2 benchmark from 35% to 77.1%, more than doubling the correct answers compared to the Gemini 3 Pro.
In addition, the new version serves as the basis for Gemini 3 Deep Think and reinforces its focus on professional use, research, and creative projects.
Consequently, the system delivers greater accuracy in challenges that require pattern recognition and structured logical deduction.
Significant Progress in Performance Testing
According to Google, the ARC-AGI-2 measures the capability of artificial intelligence to recognize abstract patterns and solve complex problems. Thus, in this assessment, Gemini 3.1 Pro answered more than twice the number of questions compared to the previous version.
-
Air bubbles from 3 million years ago extracted from Antarctic ice reveal that the current level of CO₂ has already surpassed that of the ancient, warm world, when the sea was up to 20 meters higher and there were no factories or cars burning fuel.
-
Brazil went to China to seek money and technology to reinvent TV and bring the internet to places where fiber does not reach. The mission negotiated with the BRICS Bank led by Dilma and paved the way for a Chinese rival of Starlink to connect the Amazon.
-
Iceland is going to lower a drill five kilometers to reach rock at 400 degrees and try to extract clean energy on a scale that a common well cannot achieve.
-
At the bottom of the South Pole ice, the IceCube observatory drilled more than a kilometer and a half of ice to hunt ghost particles and test whether gravity obeys the rules of quantum physics.
Therefore, the model significantly enhances logical reasoning and solidifies its application in high-demand analytical environments. Thus, it executes tasks that involve inference, analysis, and structured organization with greater precision.
Multi-Step Automation Gains Strength
In addition to the leap in the benchmark, the model expands agile workflows. Now, it autonomously executes long and structured tasks in multiple sequential steps.
For example, the system collects data from different sources and processes this information in an integrated manner. Then, it generates complete reports and dashboards, reducing the need for constant human intervention.
Consequently, teams connect different systems and tools with greater operational efficiency.
Thus, the model organizes and executes multi-step processes with increased autonomy and greater systemic integration.

More Dynamic and Interactive Digital Creation
Another advancement involves the creation of scalable SVG animations, produced in pure code and without the use of pixels. In this way, the model generates smaller files and maintains high visual quality with greater digital efficiency.
In addition, it develops interactive 3D simulations that respond to gestures and clicks in real time.
Among the examples presented, the system created a flock of manipulatable 3D birds through hand-tracking.
In the same way, the model transformed a portfolio inspired by the book Wuthering Heights into a modern interactive page.
Moreover, it built a live aerospace dashboard with public data to demonstrate the orbit of the International Space Station.
Therefore, Gemini 3.1 Pro converts abstract ideas into practical and interactive digital solutions.
Thus, it integrates logic, automation, and digital design in a more robust and application-oriented manner for professionals.
Availability and Integration in the Google Ecosystem
Currently, Gemini 3.1 Pro is available on the Gemini App for all users.
In addition, subscribers of the Pro and Ultra plans access the model on NotebookLM in a preview phase to validate updates and enhance agile workflows.
Simultaneously, developers use the model through the Gemini API, Google AI Studio, Android Studio, and Gemini CLI.
Additionally, companies integrate the system into solutions like Vertex AI and Gemini Enterprise to expand corporate applications.
Given the performance leap and the expansion of automation and digital creation, does Gemini 3.1 Pro represent merely a gradual evolution, or does it signal a deeper change in how professionals use artificial intelligence in their daily lives?

Be the first to react!