Anthropic’s recent releases of Claude Opus 4.7 and the Mythos Preview have driven Claude models to leading positions on Humanity’s Last Exam, a rigorous 2,500-question benchmark testing expert-level reasoning across math, science, and humanities. Current top Claude variants post scores between 36% and 65% depending on the leaderboard and tool-augmented setups, outpacing OpenAI’s GPT-5 series and Google’s Gemini 3.1 Pro in several evaluations. Traders focus on whether further scaling, improved chain-of-thought techniques, or an imminent Claude 5 preview can push any public model across specific thresholds before the June 30 cutoff. Competitive pressure from rapid frontier releases and ongoing benchmark updates at major AI events could shift results quickly, while evaluation variations in tool use and reasoning effort introduce uncertainty for near-term outcomes.