On February 20, Google released its latest artificial intelligence (AI) model, Gemini 3.1 Pro, which significantly enhances reasoning capabilities.
Designed for tackling complex tasks that demand advanced reasoning, this model excels in visualizing intricate concepts and synthesizing diverse data for in-depth analysis, making it ideal for challenges that go beyond simple question-answering.
In the ARC-AGI-2 benchmark, which assesses a model’s ability to navigate novel logical patterns, Gemini 3.1 Pro achieved an impressive score of 77.1%. This marks a substantial improvement, more than doubling the performance of its predecessor, Gemini 3 Pro, which scored 31.1%. Notably, it outperformed OpenAI GPT-5.2 (52.9%) and Anthropic Opus 4.6 (68.8%).
In the SWE-Bench Verified coding proficiency benchmark, Gemini 3.1 Pro scored 80.6%, nearly matching the 80.8% score of Opus 4.6, demonstrating top-tier capabilities. For the ultra-challenging Humanity’s Last Exam (HLE), it scored 44.4% without using external tools, surpassing both GPT-5.2 (34.5%) and Opus 4.6 (40%).
Following these strong benchmark performances, Gemini 3.1 Pro scored 57 in the Intel Corporation Regions Index category from the global AI assessment organization Artificial Analysis, positioning it as the world’s leading AI model.
Starting February 20, Gemini 3.1 Pro will be accessible to subscribers of the AI Pro and AI Ultra premium plans.
