Claude Fable 5.1 vs GPT 6 Astra: I gave both the same 5 briefs, ran their prompts through 5 image and video models, and measured every result.