Recent Anthropic model iterations, including Claude Opus 4.6 and variants with extended thinking and tool use, have delivered frontier scores on Humanity’s Last Exam (HLE), a 2,500-question multidisciplinary benchmark of expert-level problems, reaching 34-53% accuracy depending on configuration. These results position Claude competitively against OpenAI’s GPT-5 series and Google’s Gemini previews amid rapid iteration cycles. Trader focus centers on whether further internal scaling, prompt optimizations, or a new release before June 30 can push scores past key thresholds like 45%. Historical patterns show frontier large language models improving several percentage points monthly on such benchmarks when new capabilities ship, though exact timelines remain uncertain.