OpenAI’s GPT-6 Astra currently trails Anthropic’s Claude Opus 5.5 on independent Humanity’s Last Exam leaderboards, posting around 54.7–57% on standard evaluations versus Claude’s 61%+ as of early October 2026, yet a September 22 HLE-Diamond release showed the OpenAI model reaching 82.9% with tools, shifting trader sentiment toward stronger year-end gains. The benchmark’s 2,500 expert-authored questions test graduate-level reasoning across disciplines, and resolution hinges on official HLE Accuracy figures by December 31. Competitive pressure from Anthropic and Google releases, combined with OpenAI’s history of rapid iteration on large language models, keeps the 55%+ threshold at high implied probability while leaving room for upside on 60%+ or higher if new model variants or tool-augmented evaluations land before year-end.