New Benchmark Pits Human Writers Against 24 LLMs Across 475 Prompts, Results Show That Only Frontier Models Barely Edge Past Amateur Individuals, Showcasing A Major Skill Gap