Meta’s Muse Spark series currently trails frontier leaders like Anthropic’s Claude Opus 5.5 and Fable 5.1 on Humanity’s Last Exam (HLE), a 2,500-question expert benchmark testing advanced reasoning across sciences and humanities. Muse Spark 1.1/1.3 variants post scores in the 49–62% range across Artificial Analysis, BenchLM, and related leaderboards as of mid-October 2026, versus Claude models above 59–65% and occasional tool-augmented peaks near 73%. Rapid HLE gains industry-wide—roughly 30 points over the prior year—reflect scaling, post-training improvements, and tool use, but Meta has not yet matched the latest closed-model leaps from OpenAI or Anthropic. With only weeks remaining before the December 31 resolution, trader focus centers on whether Meta can ship a stronger Muse Spark iteration or Llama successor capable of closing the gap on the official leaderboard at agi.safe.ai.