Astra vs Fable 5.1: Which AI Builds Better Websites?

Summary

The video opens with Igor exploring how modern AI models perform when tasked with building complete websites and interactive front-end applications from scratch. To determine the strongest model for web creation, Igor built a blind benchmark comparing Anthropic's Claude Fable 5, Claude Fable 5.1, and OpenAI's GPT-6 Astra. The test suite consisted of 50 distinct prompts spanning ten categories, including dashboards, data visualizations, experimental interfaces, playable games, generative art, and marketing storefronts. Each test used frozen one-shot generation directly through model APIs without manual repair, prompt refinement, or caching.

Evaluation followed three distinct rubrics. First, Igor performed a blind human preference review, judging aesthetics, structure, and user experience side by side. Second, an automated AI judge scored the visual design quality on factors like hierarchy, rhythm, color, and SVG craft. Third, a functionality audit verified whether all prompted features were present and operational in the browser.

The first contest evaluated Claude Fable 5 against the newer Claude Fable 5.1. In the human preference test, Fable 5.1 won 30 decisions, Fable 5 took 17, and three resulted in ties. The AI visual judge rated Fable 5.1 the winner in 40 cases, giving Fable 5 only seven wins and three ties. Functionality testing resulted in a tie, with both Anthropic models successfully delivering working code on 47 of the 50 builds. Fable 5.1 showed clear improvements in margin spacing, typography hierarchy, and SVG vector rendering across marketing sites and utility layouts.

The benchmark then compared Claude Fable 5.1 against GPT-6 Astra. In the human blind choice test, GPT-6 Astra won 35 matchups, representing 70 percent of all tests, while Fable 5.1 won 15. In visual evaluation, the AI judge initially awarded 47 wins to Astra, zero to Fable 5.1, and three ties. To eliminate potential OpenAI bias, Igor re-ran the visual evaluation using Fable 5.1 as the judge, which still awarded 48 wins to Astra and only two to Fable 5.1. In functional testing, Astra passed 48 builds, while Fable 5.1 passed 47.

Category breakdowns highlighted distinct model personalities. GPT-6 Astra swept dashboards with five wins to zero, producing clean, SaaS-like control rooms and global map visualizations. Astra also dominated playable games, generating detailed dungeon crawlers with distinct entity sprites, and swept narrative editorial sites five to zero. Fable 5.1 showed distinct strengths in quirky, artistic formats, winning four out of five experimental interface tests, such as terminal simulations, and three out of five audio tools.

Finally, Igor analyzed total API costs across the 50 builds. GPT-6 Astra was the cheapest at 20.64 dollars total, or roughly 41.3 cents per site. Fable 5 cost 21.24 dollars total, or 42.5 cents per site. Fable 5.1 was the most expensive at 29.15 dollars total, averaging 58.3 cents per site.