Tech • AI • Robotics
I Tested Fable 5.1 vs Fable 5 vs Opus 5 (Cost/Speed/Design)
7/10TL;DR
A side-by-side test of leading coding models found that Claude Fable 5.1 High produced the strongest animated SpaceX landing page while costing less than several weaker alternatives, suggesting the newest model’s value depends more on effort level than hype.
KEY POINTS
What was tested
Four coding models were compared on the same task: generating an animated SpaceX Falcon 9 landing page from one prompt using five identical AI-made video clips. The lineup included Claude Fable 5.1, Claude Fable 5, Opus 5, and Codex 5.6 Soul, with low, high, and max effort settings used where available. The comparison focused on visual quality, runtime, and session cost.
Why Fable 5.1 is billed as cheaper
The pricing claim around Fable 5.1 centers on caching, which sharply lowers the price of reused input context. Standard input pricing was described as $10 per million tokens, while cached input on Fable 5.1 dropped to $0.25 per million, versus $1 per million on Fable 5. That is the basis for the widely cited 75% caching discount, though the test aimed to see whether those savings hold up in real web-building work.
How the assets were prepared
The site used five photoreal video clips of Falcon 9 liftoff, flight, orbit, and ocean landing. Those clips were generated through Higsfield, connected as a custom MCP tool inside Claude, and routed to the Seedance 2.5 video model. Asset generation took roughly 10 to 12 minutes before page building began.
Best-looking result
The highest design score went to Claude Fable 5.1 High, rated 8/10. It stood out for clean typography, strong section layout, and better overall visual flow than rival outputs. Its measured cost was $8.78, with active generation time of about 18 minutes, making it both the top-rated design and one of the more economical Claude runs.
Surprisingly strong lower-cost options
Claude Fable 5 Low and Codex 5.6 Soul Light each scored 7/10. The Fable 5 Low output was judged simple but creative, with orange accents and effective motion, taking 5 minutes 4 seconds and costing $9.76. Codex Soul Light delivered a more static but polished real-world layout in 10 minutes 53 seconds, though no session cost was available for that model.
Max settings did not dominate
Claude Fable 5.1 Max scored 7/10, not far ahead of cheaper settings, despite requiring about 26.9 minutes. Its session cost was $10.67, which was lower than expected for a top-tier maximum-effort run. Claude Fable 5 Max also scored only 6/10 while costing $24, the highest recorded price in the test.
Opus lagged on value
Opus 5 Max scored 6/10 at $10.70 and 17 minutes 57 seconds, while Opus 5 High scored 5/10 at $11 and 22 minutes 6 seconds. Opus 5 Low performed worst overall at 2/10, with broken or missing video behavior, though it was among the cheaper runs at $6.58 and 10 minutes 16 seconds. The results suggested that lower-cost models can struggle on animation-heavy design tasks.
One unusual pricing result
The cheapest recorded Claude session was Claude Fable 5 High at $5.78, despite not being the weakest setting. It completed in just 2 minutes 11 seconds and scored 6/10. That outcome reinforced the main takeaway that model cost and output quality do not scale in a clean or predictable line.
Blind ranking exposed narrow quality gaps
A blind review found much smaller differences between outputs than model branding might suggest. In several cases, low or mid-tier settings looked close to premium generations at first glance, especially for a single-page animated site. The broad pattern was that low effort settings were less reliable, but max effort often failed to justify its extra time or price.
CONCLUSION
The test suggests that Claude Fable 5.1 High currently offers one of the strongest balances of design quality, speed, and cost for AI-generated animated websites. More broadly, the results indicate that paying for maximum effort or newer branding does not guarantee meaningfully better output.
Explain this