Igor Pogany compares the first outputs from 50 website promptsFront-end code generation uses AI to create or modify the user-interface code of websites and applications from instructions or examples. spanning 10 categories, sent directly through model APIs without extra context or iterative refinement. His evaluation combines personal blind preferences with AI judgmentsA large language model as a judge is an evaluation method in which a language model scores, compares or critiques another system's output using stated criteria. of visual quality and functionality. He explicitly describes the exercise as a test of his own requirements and taste, rather than a scientifically validated benchmark.
Igor Pogany reports choosing GPT-6 Astra over Fable 5.1 in 35 of the 50 comparisons. His initial AI visual judge favors Astra 47 times, with three ties; a repeat using Fable 5.1 as judge favors Astra 48 times and Fable 5.1 twice. The stronger AI preference does not erase his own preference for Fable 5.1 in 15 cases, and the figures describe this particular prompt set and judging procedure.
Igor Pogany characterizes Astra outputs as more polished and oriented toward professional applications, with stronger SVG illustrations, dashboards and game visuals in his examples. He prefers Fable 5.1 for audio and music interfaces and experimental designs, where a less corporate style can suit the task. These are observations about default outputs: he notes that examples, style rules and more specific instructions could change the results.
Igor Pogany reports a narrow functionality difference, with Astra passing 48 of 50 checks and Fable 5.1 passing 47. The failures occur on different tasks: Astra struggles with a browser terminal example that Fable 5.1 completes, while Fable 5.1 has a non-working drawing interface that works in the Astra version. The comparison separates attractive presentation from whether the requested interactions work.
Igor Pogany reports spending $20.64 on the 50 Astra generations and $29.15 on the Fable 5.1 set, using direct API requests without cachingPrompt caching lets an AI service reuse previously processed prompt content so repeated context can cost less and run faster. or batch processing. He also reports preferring Fable 5.1 over Fable 5 in 30 of 50 earlier blind comparisons, versus 17 wins for Fable 5 and three ties. The costs and preferences are results from his experiment, not fixed per-website prices or a guarantee about refined production projects.
Watch on YouTube
