Meta’s frontier models, including the Muse Spark series, have driven recent gains on Humanity’s Last Exam (HLE), a 2,500-question multimodal benchmark of expert academic knowledge and reasoning that remains far from saturated. Muse Spark 1.1 recently posted 62.1% in verified evaluations, placing it among the top five overall and within a few points of Anthropic’s leading Claude variants at 64–65%. This reflects Meta’s accelerated scaling of proprietary systems alongside continued Llama open-weight releases, competitive pressure from closed labs, and improvements in reasoning and tool use. Traders are watching for additional Muse iterations or capability jumps before year-end that could lift Meta’s peak 2026 score, though persistent gaps in calibration and complex multi-step problems continue to favor established leaders.