Meta’s Muse Spark series, particularly versions 1.1 and 1.3, has driven recent trader focus by posting HLE scores in the 48–62% range across public leaderboards as of September 2026, placing them competitive with but typically behind Anthropic’s Claude Fable 5.1 and Opus 5 models that lead at 59–65%. The 2,500-question expert-crafted benchmark tests frontier reasoning across mathematics, physics, and other domains where closed-ended accuracy remains challenging even for advanced large language models. Sentiment centers on whether Meta can iterate its multimodal and reasoning capabilities fast enough before year-end to clear the 55% or 60% thresholds, with upcoming model releases and benchmark protocol variations serving as key swing factors amid rapid frontier progress.