No model will reach 75 percent on Humanity’s Last Exam before December 31, 2026, leaving Claude Mythos 5’s July 14, 2026 BenchLM listing of 64.5% achieved with tool use as the highest tracked score. The leaderboard will keep Muse Spark 1.1 at 62.1% and GPT-5.4 Pro at 58.7% while fourteen models rem…