Google has introduced Gemini 3.1 Pro, a preview AI model designed to handle complex multi-step reasoning and agentic workflows.
SAN FRANCISCO: Google has unveiled Gemini 3.1 Pro, describing it as a significant upgrade that more than doubles the reasoning performance of its predecessor. The new artificial intelligence model is designed to move beyond simple text generation and into advanced, autonomous task execution across industries.
Released in preview, the model introduces what Google calls “agentic” workflows. These systems are built to handle complex, multi-step problems with minimal human intervention. According to the company, the technology is particularly suited to software engineering, financial modelling and academic research, where layered reasoning and structured logic are essential.
Google said internal benchmarks show strong gains in analytical performance. On ARC-AGI-2, a demanding test that measures a model’s ability to solve entirely new logic patterns, the system scored 77.1 percent. This marks more than double the score achieved by the earlier 3.0 Pro version and narrows the gap with the recently introduced Deep Think specialised mode.
A key highlight of Gemini 3.1 Pro is its ability to generate website-ready animated Scalable Vector Graphics directly from a text prompt. This allows developers and designers to create interactive visual assets without manual coding, potentially streamlining digital production workflows.
The launch builds on the Gemini 3 architecture introduced last November. Google says the update represents a broader push to create AI systems capable of reasoning through unfamiliar problems rather than relying solely on pattern recognition from training data.
Industry analysts view Gemini 3.1 Pro as part of an intensifying race among major technology firms to develop more capable and autonomous AI platforms for enterprise and research use.
