Google announces Gemini 3.1 Pro, says it’s better at complex problem-solving

In a continued effort to innovate in artificial intelligence, Google has launched the Gemini 3.1 Pro model, following the release of Gemini 3 just last month. This latest iteration, now available in preview for both developers and consumers, promises enhanced problem-solving and reasoning capabilities. Last week, Google also announced advancements to its Deep Think tool, revealing that the foundational intelligence behind these enhancements comes from the new Gemini 3.1 Pro model. As with previous releases, Google has shared a series of benchmarks that highlight incremental improvements over its predecessor. Notably, in the Humanity’s Last Exam, which assesses advanced domain-specific knowledge, Gemini 3.1 Pro achieved a remarkable score of 44.4 percent. This is a significant increase from Gemini 3's 37.5 percent and surpasses OpenAI’s GPT 5.2, which scored 34.5 percent. Additionally, Google highlighted Gemini 3.1 Pro's performance in the challenging ARC-AGI-2 evaluation, which presents unique logic puzzles not specifically trainable for AI. The previous model, Gemini 3, scored only 31.1 percent, while the new version more than doubled that performance to an impressive 77.1 percent. Despite these achievements, Google faces stiff competition. In the Arena leaderboard, a community-driven ranking system, Gemini 3.1 Pro is slightly behind Claude Opus 4.6 by four points for text generation, and other models such as Opus 4.6, Opus 4.5, and GPT 5.2 High maintain a lead in coding tasks. It’s essential to note that the Arena leaderboard's rankings are influenced by user preferences, which can sometimes favor outputs that appear correct rather than those that are factually accurate.