Google's late-September release of Gemini 4 Argon High propelled it to the top of the Text Arena Math leaderboard by early October, with leading scores around 1530–1536 that edged out Anthropic's top Claude Opus and Fable variants. This shift, combined with strong results across related categories like coding and instruction following, has narrowed the gap and fueled trader interest despite Anthropic's 56% implied probability. The race remains tight because community-driven Elo-style voting on arena.ai can fluctuate with new model submissions, and several labs including Meta, Moonshot, and Z.ai maintain competitive entries within 30–40 points. With resolution just weeks away, any additional frontier releases or benchmark updates before month-end represent the main swing factors for final standings.