AI Research Gemini 3.1 Powers AI Research Agents That Outperform Human Analysis Gemini 3.1 cut hallucinations by 38 percentage points and hit 77.1% on ARC-AGI-2—but its research agents raise a uncomfortable question about human analysts. AITechBrief EditorApril 28, 2026