Google’s Gemini Pro’s latest iteration significantly boosts its reasoning capabilities, achieving over double the performance on the ARC-AGI-2 benchmark, according to the Google Gemini team.
Google has unveiled a sneak peek of Gemini 3.1 Pro, positioning it as a more intelligent model adept at solving highly complex problems and representing a significant leap in fundamental reasoning abilities.
Announced on February 19, Gemini 3.1 Pro is engineered for tasks demanding more than simple answers, leveraging enhanced reasoning to tackle the most formidable challenges, as stated by the Google Gemini team. This augmented intelligence finds practical use in applications such as visually explaining intricate concepts, consolidating diverse data into a unified view, and resolving issues that necessitate extensive context and strategic planning. The model is currently available for developers as a preview via the Gemini API in Google AI Studio, Gemini CLI, Google Antigravity, and Android Studio. For business clients, the model is accessible through Vertex and Gemini Enterprise. Individual users can engage with Gemini 3.1 Pro via the Gemini app and NotebookLM.
Gemini 3.1 Pro arrives after the Gemini 3.1 release from November 2025. The Gemini team highlighted that the core intelligence powering Gemini 3.1 Pro was also integral to last week’s update for Gemini 3 Deep Think, enabling it to conquer challenges across scientific, research, and engineering domains. Furthermore, the team reported that on the ARC-AGI-2 benchmark, designed to assess a model’s proficiency in deciphering novel logic patterns, Gemini 3.1 Pro achieved a verified score of 77.1%, which is more than double the reasoning prowess demonstrated by Gemini 3 Pro.
