Claude Opus 4.8 vs Fable 5
Which Claude AI model works best for design tasks

Claude Opus 4.8 vs Fable 5
Which Claude AI model works best for design tasks
With so much hype around Fable 5, it can be hard to separate facts from fiction. Some people say that Fable 5 is a groundbreaking model that will change how we build products. Others say that Fable 5 is really bad at token efficiency and uses far more tokens than its predecessors.
I think the only right way to know what is true and what is not is to test the model. And to make the testing as relevant to everyday usage as possible, I will test both Claude AI models on one of the most common design tasks: creating a landing page.
[embed]
Testing preconditions
Before showing you testing results, I want to explain my testing process and focus on a few essential aspects of it:
Page details and original prompt
I will create a landing page for a women’s fragrance (this is not a real product). The prompt outlines the main attributes of the product, product perception, and content sections I want to see on the page.
Create a premium landing page for a new women’s fragrance called J'adore
Product: a modern feminine fragrance designed for confident, elegant women. The scent combines fresh bergamot, soft jasmine, rose petals, vanilla, and warm sandalwood. The overall feeling should be luxurious, sensual, clean, and emotionally memorable.
Design direction: Create a sophisticated landing page with a high-end beauty/fashion aesthetic. Use soft neutral tones, warm beige, ivory, muted rose, and subtle gold accents. The page should feel elegant, airy, and premium, not overly decorative. Use large editorial typography, generous whitespace, smooth transitions, and beautiful product-focused visuals.
Landing page structure:
1. Hero section with a strong headline, short emotional description, fragrance bottle visual, and primary CTA.
2. Brand story section that explains the inspiration behind the fragrance.
3. Scent notes section showing top, heart, and base notes in a beautiful visual layout.
4. Product benefits section focused on long-lasting scent, elegant bottle design, and day-to-night wear.
5. Lifestyle section showing the fragrance as part of a confident woman’s daily ritual.
6. Social proof section with short customer quotes.
7. Final CTA section encouraging users to discover or purchase the fragrance.
Copy tone: Make the copy poetic but clear. Avoid generic marketing clichés. The fragrance should feel desirable, premium, and emotionally connected to identity, memory, and confidence.
UX requirements: The page should be visually striking above the fold, mobile-friendly, conversion-focused, and easy to scan. Include strong CTAs such as “Discover the Fragrance,” “Shop Now,” or “Experience the Scent.”
Output: Provide the full landing page in HTML, CSS and Javascript
Don’t use Claude Skills for this
I have a few front-end design skills installed in my Claude Code environment, but to make an honest comparison between Opus and Fable, I will ask Claude not to use the skills.
Don't use Claude Skills for this
Effort
Effort is a special setting that defines how much effort the AI model will invest in your task.

Customizing the effort level for the Fable 5 AI model.
I will use Extra High effort for both Opus 4.8 and Fable 5. According to Anthropic, Fable 5 tends to perform much better at Extra High effort.
Claude AI model scoring for different effort levels.
Quick note: You might wonder why I chose to use Extra High and not Max. The reason for that is simple: the quality of output that the AI model generates for you. You will get pretty much the same quality of output for Fable 5 when using Extra High effort, and, quite surprisingly, better quality for Opus 4.8. But what is even more important is that you save a lot of tokens.
Claude session and project directory
I will run two separate sessions, one with Opus 4.8 and another with Fable 5, so that models cannot reuse results of the previous session when building a new design. I will also choose two separate workspaces for my design (directories on the local drive) that were empty before the test.

jadore_opus_48 and jadore_fable_5 are two work directories where I expect Claude to generate design.
Opus 4.8
Here is a result that Opus 4.8 generated for the prompt mentioned above. As you can see, it’s a pretty good web page with a few nice details such as smooth animated transitions and scroll-triggered sliding effects. Perhaps the main downside of this design is that it’s very predictable: this web page is exactly what you can expect to get from an AI tool.

Landing page design by Opus 4.8
Opus 4.8 generated this page in 4 minutes, and it cost 2% of my daily usage limit for Pro plan (which is quite reasonable).
Fable 5
And here is what Fable 5 generates for me for the same prompt. Right from the first glance, it looks like a more polished and sophisticated web design. Animated transitions are also a major step up in comparison to Opus 4.8.

Landing page design by Fable 5
It took twice as much time for Fable 5 to generate this design (8 minutes), and it took 18% of my daily usage quota (which is even more reasonable because the quality of output is much better).
Side note: One interesting thing that I’ve noticed about Fable 5 is that it uses self-validation for the results it generates. You can notice in the console output that it triggers previews (captures screenshots of the web page layout it generated and uses it as an additional contextual input to refine the desing).

Refinement loop in Fable 5.
And the winner is
This time it’s a no-brainer. Fable 5. And my test will also burst the popular myth about Fable 5 token usage inefficiency. In my experience, Fable 5 is a very token-efficient model (again, considering the quality of output you get).
Want to master Claude Code?
Check out my complete guide to Claude Code, packed with highly practical insights on how you can integrate it into your design process. This guide is updated on a weekly basis, so you’ll always get the latest information about Claude Code.
메타데이터
- post_id
- 2b0a442ccdcb
- slug
- claude-opus-4-8-vs-fable-5-2b0a442ccdcb
- url
- https://uxplanet.org/claude-opus-4-8-vs-fable-5-2b0a442ccdcb
- canonical_url
- https://uxplanet.org/claude-opus-4-8-vs-fable-5-2b0a442ccdcb
- author_url
- https://medium.com/@101
- status
- ok
- fetched_at
- 2026-06-15 20:49:13