
Astra 6.2 and Fable 5.2 are AI models that can help produce everything from websites to financial workbooks. Their results differ in quality, runtime and API spending, even when they receive the same prompts. In a 12-task comparison, Fable won seven tasks, while Astra was faster and cheaper across the comparison, according to Nate Herk.
Compare how the models handled trip planning and financial workbooks, including the level of detail each provided. You can also see why Fable’s total API bill was 7.2 times higher despite identical per-token rates, and consider the limits of the test, which did not assess intensive refactoring or large codebases.
ChatGPT Astra 6.2 vs Claude Fable 5.2
TL;DR Key Takeaways :
- Fable 5.2’s performance was better in more tasks, but its total API cost was 7.2 times Astra 6.2’s; Astra was cheaper and faster across the comparison.
- Both models have the same stated API rates: $8 per million input tokens and $40 per million output tokens. Fable’s higher bills came from greater token use.
- Astra won for the website, portrait and learning guide, and its simpler presentation was preferred for projector use. Fable’s more detailed presentation was seen as better suited to reading by email.
- Fable won the video-editing rounds, virtual summit, trip planning, financial workbook and browser game, often delivering more detailed or better-suited outputs at greater cost.
- The tests did not establish performance on intensive refactoring or large codebases. Herk thinks more specific prompts and skills could bring Astra closer to Fable’s outputs for less, provided Astra calls the right skill.
Websites, Presentations and Painting
For a website built from existing brand assets, Astra produced a relatively clean page with a dynamic navigation bar. However, a scrolling element showed its third image as the visitor was moving away, and a flavor-selection image was cropped.
Fable’s site had more movement, including animated text, an orbiting ring and a coffee-and-protein swirl. Visitors could switch flavors in two places, and the flavor images fit their crops. Astra won on design and cost: its run took 12 minutes, versus 56 minutes for Fable, with estimated API costs of about $2.50 and $175, respectively.
Both models turned a mock survey into a 12-slide presentation using existing brand guidelines. Astra’s deck used the brand colors and logo, charts and relatively little text, although roughly three slides in a row repeated the same format. Fable used more words and slide layouts, with branding in the corner of each slide.
Astra’s simpler deck was preferred for presenting on a projector, while Fable’s more detailed version was seen as better suited to reading by email. Astra won on cost: its estimated API charge was about $2, compared with about $9 for Fable. Both models may be more than the task requires.
For a portrait painted in Canva from a reference image, Astra produced a more realistic result, while Fable’s painting had visible marks and gaps. Codex, working with Astra, got around Canva’s stroke limit by exporting and reimporting the painting before continuing. Fable repeatedly hit the limit and stopped. Astra won after running for an hour and a half at an API cost of about $7.50. Fable ran for three hours and spent $63.
Video Editing and a Virtual Summit
The models were asked to turn event footage into an energetic 60-second recap. Neither output contained real speech. Fable won, although both outputs were judged poor; the prompting may have caused the result. Astra’s run cost about $4 and Fable’s $15.
For an Instagram reel, Fable won on sound design and B-roll selection. It used more tokens and cost about four times as much as Astra. Nate Herk notes that the difference was worthwhile for that output.
The long-form YouTube editing task exposed a difference in how the models used existing editing instructions. Astra added B-roll but used an older skill rather than the intended style. Leaving that old skill in place may have contributed to the result, even though recent context and folders used the intended one. Its output cost $7.
Fable used the correct skill and produced the requested colors, animations and branded introduction, winning the round. Its run took more than five hours, longer than the usual workflow using Opus 5.5.
For a playable virtual summit made from event recordings and assets, Fable created an agenda with real session times, navigation to stages, playable recordings, booths and speaker information. Its mouse controls were inverted, and some design elements looked cheap. Astra’s world had shadows and better-looking avatars, but its logo was slightly wrong. Fable won for the overall experience, although the runs took similar time and Fable cost more.
Advance your skills in Astra 6.2 and Fable 5.2 by reading more of our detailed content.
- Fable 5.2 Drops False Positives by 55% but Downgrades to Opus 4.8
- ChatGPT 6 Officially Launches as Astra: Beats GPT-5.6 Sol with 47% Faster Speed
- New Fable 5.2 AI Built a Complete Operating System in 8 Days
- Anthropic’s Fable 5.2 Codes a Gears of War Clone for About $15,000
- Claude Mythos 5 and Fable 5 Officially Launch
Learning Guides and Trip Planning
Given a full tutorial, both models produced visual learning guides with timestamps and annotated screenshots. Fable’s HTML guide played the full tutorial and individual clips, cropped and labeled screenshots, and tracked completed items with a “got it” control.
Astra’s guide opened the tutorial in a full-screen view and used smaller, less detailed annotations. It won the subjective comparison for its cleaner layout and lower cost: about $3, versus $17 for Fable.
For a trip planned from a vague request, the models independently chose a similar route. Astra allocated nights to stops but did not give specific hotels or flights to book, so those details would require another prompt.
Fable’s planner supplied schedules, links to check and driving legs on a map that zoomed in at stops. It won because those details mattered more for trip planning than Astra’s more attractive design. Fable cost $18.60 for the task, compared with $2.70 for Astra.
Carousels, Financial Models and Games
Neither model followed the established format when asked to turn an article into an eight-slide Instagram carousel, despite existing brand guidelines and a skill for the account. The expected format included consistent branding, a bottom logo and a standard final slide. Astra won as the lesser of two unsatisfactory results: its design was less overwhelming and its run was cheaper.
For financial data, both models made simple interactive revenue dashboards, with Astra’s appearing more visual. Their Excel workbooks differed more: Astra provided an overview, history, forecast and minimal charts, while Fable added more graphs and numbers and a more structured drivers section.
The parts of Astra’s workbook checked in testing had formulas. Fable’s also had formulas that appeared to extend through most of the document and reference other sheets. Fable won on the detail and reasoning behind its numbers and visualizations, although whether that extra detail is useful depends on the reader. Its run cost $15, versus $2.50 for Astra, and took about two to almost three times as long.
Both browser games used the name Brief Escape. Astra’s was two-dimensional and allowed walking, jumping and limited angle changes. Fable’s was 3D, with a cat, a checkpoint and the ability to fly. It won for the gaming experience at a cost of $19, compared with $3 for Astra.
API Pricing and Use-Case Limits
The models have the same stated API rates: $8 per million input tokens and $40 per million output tokens. The larger Fable bills in this comparison came from greater token use, rather than a higher per-token rate.
Nate Herk favored Fable 5.2 in more tasks, but its total API cost was 7.2 times Astra 6.2’s. Fable’s preferred outputs came at a higher cost, while Astra was cheaper and faster across the comparison. The tests did not establish how either model performs on intensive refactoring or large codebases. Herk thinks more specific prompts and skills could bring Astra closer to Fable’s outputs for less, provided Astra calls the right skill.
Media Credit: Nate Herk | AI Automation
Disclosure: Some of our articles include affiliate links. If you buy something through one of these links, Geeky Gadgets may earn an affiliate commission. Learn about our Disclosure Policy.