xAI’s Grok models currently trail on Humanity’s Last Exam (HLE), a 2,500-question expert benchmark of graduate-level knowledge and multi-step reasoning, with Grok 4.6 at 42.9% and 4.5 at 42.7% versus Claude Fable 5.1 leading at 55–65% on recent leaderboards. Trader sentiment centers on whether xAI’s aggressive scaling—evident in the ongoing 2.5-trillion-parameter Grok 4.8 training run and prior jumps like Grok 4 Heavy’s tool-augmented gains—can close the gap by year-end through larger models, refined reinforcement learning, and faster iteration cycles. Anthropic and OpenAI maintain leads via specialized reasoning variants, while HLE’s living updates and historical rapid progress (frontier rising from single digits to mid-40s% in under two years) create room for surprises. Key catalysts include the imminent 4.8 release, any new verified submissions before December 31, and potential tool-augmented or adaptive modes that could shift official leaderboard positions.