Recent Anthropic releases of Claude Fable 5 and Mythos 5, launched June 9, 2026, have propelled the models to leading Humanity’s Last Exam scores of 53–64.5% depending on tool use and configuration, outpacing Gemini 3.1 Pro and GPT-5 variants on this expert-level 2,500-question benchmark spanning advanced science, math, and reasoning. These gains stem from enhanced adaptive reasoning, agentic capabilities, and selective fallback mechanisms, establishing Anthropic’s edge in frontier large language model performance. With June 30 just weeks away and no announced follow-up updates, trader consensus reflects the durability of these verified results on public leaderboards, tempered by typical variance in independent evaluations and potential refinements in testing protocols.