Google announces Gemini 3.1 Pro, says it’s better at complex problem-solving (arstechnica.com)

Google has unveiled Gemini 3.1 Pro, an upgraded version of its AI model aimed at enhancing complex problem-solving and reasoning abilities. This release follows the November launch of Gemini 3 and is now available in preview for developers and consumers. The latest model showcases significant improvements, especially in scoring 44.4 percent on the Humanity’s Last Exam, which evaluates advanced domain knowledge. This is a notable jump from Gemini 3’s 37.5 percent and positions it ahead of OpenAI's GPT 5.2, which scored 34.5 percent. Furthermore, in the ARC-AGI-2 benchmark, Gemini 3.1 Pro excelled by achieving 77.1 percent, a substantial improvement over its predecessor's 31.1 percent, showcasing its enhanced ability to address novel logic problems. While Google’s latest model claims notable advancements, it has not secured the top positions on the Arena leaderboard for text and code generation, where competing models like Claude Opus 4.6 and GPT 5.2 High still lead. Nevertheless, Google continues to assert its progress in AI development, emphasizing that the improvements in Gemini 3.1 Pro could have significant implications for applications requiring deeper analytical capabilities, such as scientific research, complex data analysis, and advanced logic problem-solving.