Meta’s Muse Spark 1.1 series, released in April 2026 with subsequent refinements, has driven recent gains by posting 58–62% accuracy on Humanity’s Last Exam depending on the evaluation setup and “contemplating” or high-reasoning modes. These scores place Meta models within striking distance of Anthropic’s Claude Mythos and Fable variants while outperforming earlier Llama releases and offering competitive pricing for frontier performance. Ongoing internal scaling efforts, combined with the benchmark’s 2,500 expert-level questions across science and humanities, keep trader attention focused on whether Meta can close the gap or launch a follow-on model before year-end.