Recent progress in large language models and reasoning systems has pushed AI closer to gold-medal thresholds on Olympiad problems, with several frontier models demonstrating strong results on 2025-style benchmarks and unofficial 2026 problem sets. However, traders assign “No” the lead at 64.5 percent because official IMO grading requires fixed submission windows, no human steering, and formal scoring by organizers—criteria that remain unproven at scale for most labs heading into the July contest. Variability in visual reasoning, proof rigor, and cross-problem consistency continues to temper confidence, while historical patterns show model performance can fluctuate sharply on novel problem distributions. Key near-term catalysts include any verified official scores from the Shanghai event and subsequent developer statements on training timelines.