Meta’s Muse Spark series, particularly the 1.1 variant, currently posts the company’s strongest Humanity’s Last Exam results near 62% on select leaderboards with tools, trailing Anthropic’s Claude Opus 5.5 (67.7%) and Fable 5.1 (65%) while remaining competitive with several GPT-5 and Gemini entries. This positioning reflects Meta’s emphasis on open-weight releases and efficient scaling rather than closed frontier capability races, alongside steady benchmark gains of roughly 30 points over the prior year across the industry. Traders are watching for potential late-2026 model updates or tool-augmented evaluations that could push Meta past the 65% threshold before year-end, though historical release cadence and the benchmark’s resistance to rapid saturation suggest meaningful gaps may persist against leading labs.