xAI’s Grok models currently lag on Epoch AI’s FrontierMath benchmark, with Grok 4 variants at roughly 12–14% on comparable tiers while leading large language models reach over 50% as of late May 2026. The market focuses on whether any Grok version can hit the 25% or 30% threshold by June 30—a narrow window driven by the benchmark’s hard, contamination-resistant research-level problems that test advanced mathematical reasoning beyond standard tool use. Recent Grok 4.20 updates prioritized multi-agent systems and real-time data over specialized math scaling, with no official xAI announcements confirming imminent FrontierMath evaluations. Traders weigh xAI’s substantial compute and competitive pressure against the short runway and historical difficulty of rapid benchmark jumps in this frontier AI domain.