Google Launches Gemini 3.1 Pro With Double the Reasoning Power
Google introduces Gemini 3.1 Pro, a mid-cycle upgrade that brings advanced reasoning capabilities from Deep Think to a wider audience. The new model scores 77.1% on the ARC-AGI-2 benchmark, doubling the performance of its predecessor.
Google releases Gemini 3.1 Pro, marking the company's first point-one mid-cycle update and shifting away from the traditional point-five increment. This new model focuses specifically on complex problem-solving, targeting situations where a simple answer falls short of user needs.
The upgraded core intelligence leverages the same advanced reasoning technology that recently debuted in Gemini 3 Deep Think. This translates to an impressive ARC-AGI-2 benchmark score of 77.1%, which represents more than double the reasoning performance of the standard Gemini 3 Pro model.
Gemini 3.1 Pro rolls out immediately in the Gemini app and NotebookLM for AI Pro and Ultra subscribers, while developers access it through the Gemini API, Vertex AI, and other Google developer tools. The model currently launches in preview so Google can validate the updates and refine ambitious agentic workflows before general availability.