Pat Simmons gives GPT-6 Astra and Fable 5.1 the same one-shot promptsOne-shot prompting asks an AI system to complete a task from one initial instruction without iterative correction. for three demanding builds: a macOS iPod with generated music, an interactive D-Day reconstruction and a playable browser shooterGame generation uses AI to create or assemble playable assets, scenes, rules, audio and interactions.. Both systems must deliver without iterative corrections before the results are tested directlyReal-world AI evaluation measures model or agent behavior in genuine deployment conditions where actions encounter actual users, systems, constraints, and consequences..
Fable produces a polished iPod with useful sound effects and navigation, while Astra's version looks more realistic and detailed. In the D-Day project, Astra delivers clearer navigation, stronger terrain and water detail and a more convincing interactive reconstruction. Its browser shooter also has more realistic characters and a recognizable map, though aiming, reloading and damage behavior still need workGameplay testing checks whether a game's controls, rules, feedback and player experience work as intended..
Simmons concludes that Astra wins the comparison through creativity, visual detail and thoroughnessCoding output quality measures whether AI-generated software is correct, maintainable, complete and suitable for its intended use.. It takes longer, but its higher usage allowance and ability to keep working on large builds make it his preferred coding model in this test, while Fable remains competitive on interaction polish and speed.
Watch on YouTube
