“On a per-task basis in the Artificial Analysis benchmark, Grok 4.6 is 32% cheaper than GPT-5.6-SOL and 73% cheaper than Fable.”