GPT-5.6 Sol, Fable 5, Grok 4.5 and GLM 5.2 competed on the same math

One prompt, four frontier models, every number checked by hand. Three tied. One was quietly wrong