Compare GPT-6.2 Astra and Claude Fable 5.2 on independent benchmarks, reasoning effort, real task costs, long-context pricing ...