GPT-6 Astra vs Claude Fable 5.1: Which Builds Better Websites?
A 50-site blind benchmark tests GPT-6 Astra against Claude Fable 5.1 on one-shot website generation, visuals, and functionality.
Which AI builds better websites, GPT-6 Astra or Claude Fable 5.1?
Across a 50-site blind benchmark spanning ten categories, GPT-6 Astra beat Claude Fable 5.1 in one-shot website generation by a wide margin in human preference (35 out of 50 wins) and by an even wider margin according to AI judges (as high as 48 out of 50). Fable 5.1 still held its ground in specific categories like audio interfaces and experimental, personality-driven design. Functionality was close to a tie between the two models.
TL;DR
- GPT-6 Astra won the overall matchup, taking 35 of 50 blind human preference tests against Fable 5.1, a roughly 70% win rate.
- AI judges preferred Astra even more heavily than the human tester did, with one AI judge picking Astra 47 times out of 50 and another picking it 48 times out of 50, regardless of which model’s own family did the judging.
- Fable 5.1 clearly beat its predecessor, Fable 5, winning 30 of 50 human preference tests and 40 of 50 AI visual judgments, so Anthropic’s update was a real step up before Astra’s release.
- Functionality was nearly identical between Astra and Fable 5.1, with Astra passing 48 of 50 functional tests and Fable 5.1 passing 47 of 50, meaning the gap between the two is mostly about visual and design quality, not whether the sites actually work.
- Category matters more than overall score: Astra dominated dashboards, data visualizations, games, and narrative/editorial sites, while Fable 5.1 held an edge in audio/music interfaces and experimental, unconventional designs.
- Astra’s SVG and vector graphic generation was consistently stronger, especially visible in map recreations, game character art, and geometric illustrations.
- Fable 5.1’s default aesthetic reads as more casual and idiosyncratic, while Astra’s default output leans toward polished, SaaS-style professional design, which is a stylistic tradeoff rather than a strict quality gap.
Remy doesn't build the plumbing. It inherits it.
Other agents wire up auth, databases, models, and integrations from scratch every time you ask them to build something.
Remy ships with all of it from MindStudio — so every cycle goes into the app you actually want.
How was this benchmark actually run?
The comparison used 50 one-shot prompts spread across ten categories of websites and applications, sent directly to each model’s API with no extra context or iterative refinement. One-shot means the prompt goes in once and the first output is what gets scored, there’s no back-and-forth polishing or follow-up correction. That constraint matters: it tests what a model produces by default, not what it’s capable of with a skilled prompter guiding it through multiple rounds.
The test set intentionally moved beyond generic templates like personal portfolios or e-commerce storefronts. It included harder, more varied builds: interactive games, physics and music simulators, data dashboards, generative art tools like an ink-drawing studio, a bird-flock animation lab, a coffee configurator, and narrative/editorial sites like recipe pages and museum guides.
Every site was scored across three rubrics: personal blind human preference, an AI judge’s rating of visual quality, and an AI judge’s rating of functionality (whether every requested feature was present and worked). The human preference scoring was done blind, meaning the identity of which model produced which site was hidden during judgment, and matchups were run separately for Fable 5 vs Fable 5.1 and Fable 5.1 vs Astra.
How did Fable 5.1 compare to its predecessor, Fable 5?
Before Astra entered the picture, Fable 5.1 was tested against the earlier Fable 5, and it won clearly. Fable 5.1 took 30 of 50 blind human preference matchups versus 17 for Fable 5, with three ties. AI judges were even more decisive, favoring Fable 5.1 in 40 of 50 cases against just seven for Fable 5, again with three ties.
The biggest visible improvement was spacing and visual hierarchy. Fable 5.1’s layouts consistently gave elements more breathing room, avoided uniform sizing between headlines, buttons, and controls, and cut off content at more sensible points. SVG and vector graphic quality was also noticeably higher in the newer model. On functionality, the two versions tied exactly, meaning the upgrade was about design polish, not about fixing broken features.
Why did GPT-6 Astra win the overall matchup against Fable 5.1?
Astra’s advantage showed up most clearly in three areas: vector graphics, games, and structured data interfaces like dashboards.
Astra’s SVG generation was consistently more sophisticated, visible in things like a hand-drawn world map recreated in vector form or detailed illustrated icons. In games, the gap was described as not close at all: Astra-built games featured actual character and enemy art with visual personality, while Fable’s game outputs tended toward simpler, more abstract representations like basic icons standing in for players and enemies.
Dashboards and data visualization sites were another clean sweep for Astra. Layouts felt more resolved, spacing was more considered, and interactive elements like expandable panels worked more intuitively. The same pattern held for narrative and editorial-style sites, spatial storytelling pages built around a specific theme or product.
Plans first. Then code.
Remy writes the spec, manages the build, and ships the app.
Stylistically, Astra’s default output leans toward a professional, SaaS-like look: headers, top navigation, consistent typographic hierarchy. Fable 5.1’s default output was described as more casual and idiosyncratic, sometimes closer to something a less experienced designer would put together by hand. Neither is objectively wrong, it’s a difference in default taste, and it can be steered with more explicit style guidance in the prompt. But out of the box, on a single unguided prompt, Astra’s professional polish scored higher across most categories.
Where does Fable 5.1 still win?
Fable 5.1 held a real advantage in two categories: audio and music interfaces, and experimental or unconventional interface designs. These are the categories that reward a more expressive, less conventional visual approach rather than clean structure. A kinetic poem site and other animation-heavy experimental builds were cited as places where Fable 5.1’s default personality fit the assignment better than Astra’s more corporate-leaning interpretation of the same prompt.
This suggests the choice between the two models isn’t purely about overall quality. It’s about matching the model’s default aesthetic to what the project actually needs. A funky one-off art piece or music tool may benefit from Fable 5.1’s looser style, while a dashboard, admin panel, or game will likely come out stronger from Astra on the first try.
Did functionality differ between the two models?
Barely. Astra passed 48 of 50 functional tests, Fable 5.1 passed 47 of 50, essentially a wash. Both models occasionally produced sites where a feature didn’t fully work: Astra struggled with a topic bubble explorer and a simulated personal terminal app, both fairly complex interactive builds, while Fable 5.1 had issues with an ink-drawing studio that failed to actually draw, a RAG learning app, and an ocean restoration report site. None of these failures were widespread, and both models generally deliver working one-shot builds for most prompt types. The real differentiator between them is visual and design quality, not whether the code runs.
Is one model clearly worth choosing over the other?
For most general-purpose website generation, especially anything visual, data-driven, or game-like, Astra appears to be the stronger default choice based on this benchmark’s human and AI scoring. But Fable 5.1 isn’t obsolete: its more expressive, less templated aesthetic makes it a better fit for creative, experimental, or audio-driven projects where personality matters more than polish. Builders who care about one-shot output quality without heavy prompt engineering should treat category as the deciding factor rather than picking one model as a universal default.
Frequently Asked Questions
What does “one-shot” mean in this website generation benchmark?
It means each prompt was submitted once, directly to the model’s API, and the very first output was scored. There was no iterative refinement, follow-up correction, or additional context provided, so the results reflect default model behavior rather than what’s achievable with extended prompting or multi-turn editing.
Did AI judges and human judgment agree on which model was better?
Broadly yes, but AI judges were far more one-sided. The human tester preferred Astra in 35 of 50 cases, while AI judges preferred Astra in 47 or 48 out of 50 cases depending on which model did the judging, with almost no wins for Fable 5.1 in the AI-scored visual rubric.
Is Fable 5.1 a meaningful upgrade over Fable 5?
Seven tools to build an app. Or just Remy.
Editor, preview, AI agents, deploy — all in one tab. Nothing to install.
Yes. Fable 5.1 won the majority of blind comparisons against Fable 5 on both human preference and AI visual scoring, driven mainly by better spacing, visual hierarchy, and SVG quality. Functionality between the two versions was identical.
Which categories favor Fable 5.1 over Astra?
Audio and music interfaces, along with experimental or unconventional creative interfaces, were the categories where Fable 5.1’s more casual, idiosyncratic style outperformed Astra’s more polished, professional default look.
Does either model reliably produce fully working websites on the first try?
Mostly yes. Both models passed nearly all functional tests (48 of 50 for Astra, 47 of 50 for Fable 5.1), with only a handful of complex interactive builds, like terminal simulators or drawing tools, showing issues on the first generation.
