Meta’s Muse Spark and Muse Wonder models currently post the strongest results among its releases on Humanity’s Last Exam, with scores reaching the low-to-mid 60% range on tool-augmented or specific leaderboard variants and the low-to-mid 50% range on standard Artificial Analysis and Scale evaluations. These place Meta models competitively but behind Anthropic’s Claude Opus 5.5 (61%+ without tools, 67%+ with tools) and OpenAI’s GPT-6 Astra variants. Rapid benchmark gains across frontier labs in 2026, including refined subsets like HLE-Diamond, have raised expectations for further Meta releases or scaling improvements before year-end, though product timelines and exact resolution criteria on the official leaderboard introduce uncertainty for thresholds above 60%.