Meta’s Muse Spark series, particularly the 1.1 and 1.3 variants, has driven trader sentiment by posting frontier scores of 48–62% on Humanity’s Last Exam, a 2,500-question expert-crafted benchmark spanning mathematics, physics, and other domains that resists saturation. These results place Meta models within a few points of Anthropic’s leading Claude Fable 5.1 and Opus 5 on certain leaderboards while outperforming several GPT-5 and Gemini configurations. Momentum stems from expanded training infrastructure, adaptive reasoning techniques, and iterative post-training refinements that have narrowed the gap with closed-source leaders. Key remaining catalysts include potential fall 2026 model drops, further scaling gains, and any HLE updates with new held-out questions before year-end resolution.