Gemini 3.1 Pro: Google deploys a model more comfortable with complex problems

AI models follow each other and start to look alike. After updating Gemini 3 Deep Think last week to "solve modern problems in the fields of science, research, and engineering", the web giant announced the deployment of Gemini 3.1 Pro which integrates the core improvements of Deep Think.

Google is putting oil on its AI engine

The model will appear (unless it already has) in the NotebookLM and Gemini apps for Google AI Pro and Ultra subscribers, for businesses and developers via dedicated Gemini platforms. The focus is on reasoning; Google presents 3.1 Pro as a smarter and more robust foundation for solving complex problems.

To give an idea of the capabilities of this new model, the results of the ARC-AGI-2 benchmark are promising. It is a test that measures a model's ability to understand and solve logical problems it has never seen before, without relying on examples it might have in memory. Gemini 3.1 Pro achieves a score of 77.1%, which is more than double the reasoning performance displayed by Gemini 3 Pro.

Google provides a handful of examples of the model's capabilities: creating animated SVG images, generating code for an International Space Station tracking tool, manipulating a point cloud, transforming website themes into functional code… This model is somewhat equivalent to a "speed bump" or a minor update for Android or iOS.

👉 Follow real-time tech news: add 01net to your Google sources, and subscribe to our WhatsApp channel.

Source : Google