Google just released the Gemini 3.1 Pro model, and what excites me the most are its scores on these three benchmark tests: ARC-AGI-2: Reasoning intelligence improved by 2.5x 31% → 77%, no need to say much about this test—it’s currently the most hardcore IQ test, purely based on logic, no way to memorize answers. BrowseComp: Deep search capability improved by 1.5x 59% → 86%, searches for things hidden in hard-to-find corners, not just simple Google searches, also tests the ability to self-correct when results aren’t found. APEX-Agents: Long tasks/professional tasks improved by 2x 18% → 36%, mainly evaluates long-term professional tasks, especially career-related ones like investment banking analyst, lawyer, consultant—requires reading reports and documents. My Hakimi is getting smarter and smarter .