Recent releases of OpenAI's GPT-6 series and Anthropic's Claude Opus 5 models have driven top MathArena scores into the low-to-mid 80s as of late October 2026, with GPT-6.1 Sol leading at roughly 84% expected performance across recent competitions. These gains stem from improved chain-of-thought reasoning and tool use on uncontaminated problems from sources like AIME, HMMT, and FrontierMath. Trader sentiment reflects the tight race between OpenAI and Anthropic labs, limited remaining calendar time before December 31, and the difficulty of rapid further gains on saturated or harder tiers. New model updates or scaling announcements in November could still shift implied probabilities for ambitious score thresholds.
如果 Arena.AI 排行榜(arena.ai/leaderboard/text)上的任何模型在 2026年12月31日 23:59 ET 前,于“Math”的“Leaderboard”标签页达到至少指定的 Arena Score,本市场将结算为“是”。否则,本市场将以“否”结算。 本市场将根据 https://arena.ai/leaderboard/text/math-no-style-control 上“Text Arena | Math”排行榜标签页中“Score”列的结果进行结算,且样式控制项须关闭。 本市场的结算来源是 arena.ai/leaderboard/text 上的 Chatbot Arena LLM 排行榜。如果检查时无法使用此结算来源,本市场将保持开放,直到排行榜重新上线,并以重新上线后的首次检查结果结算。如果该来源永久不可用,本市场将结算为“否”。
机器翻译。如有争议,以原文为准。