Gemini 4 Hydrogen Shows Diverging Results Across Reasoning Levels
AI summary
On ARC-AGI-2, Gemini 4 Hydrogen outperformed Gemini 3.8 Flash at the medium and high reasoning levels, scoring 89.2% and 90.0%, respectively, versus 82.9% and 89.2%. However, at the low level, it trailed the latter with 65.8% versus 77.5%. Gemini 4 Hydrogen also adds an xhigh reasoning level not available in Gemini 3.8 Flash.
Body
@GoogleDeepMind On ARC-AGI-2, Gemini 4 Hydrogen scored higher than Gemini 3.8 Flash at medium (89.2% vs 82.9%) and high (90.0% vs 89.2%), though trailed at low reasoning (65.8% vs 77.5%).
Gemini 4 Hydrogen also introduced a xhigh reasoning level that was not available for Gemini 3.8 Flash. https://t.co/yEewYk1IuM
AI Trends