Summary
Creator Nate Herk spent a full day running GPT-5.6 Soul and Fable 5 (Claude) side by side through a series of real-world coding and creative tasks to build an honest picture of which model delivers more value in day-to-day work. Both models were given identical prompts across multiple challenges, starting with a fully playable open-world browser bike game and an ‘impossible’ interactive scrolling website, with results evaluated on output quality, completion time, and cost.
The results reveal a clear cost-quality tradeoff. Fable 5 produced a noticeably more immersive 3D game — at $14.22 and roughly 90,000 output tokens — while GPT-5.6 Soul completed a similar task for $4.50 using only 31,000 tokens. Across creative canvas challenges (generative art tools, interactive maps, typing games, underwater scroll stories), the pattern held: Soul is dramatically more token-efficient, which sometimes means faster, cheaper, and ‘good enough,’ but occasionally means it stops short of the more elaborate output Fable produces when given room to explore.
Nate’s overall framework: GPT-5.6 Soul is the better default for structured, goal-directed coding tasks where token efficiency matters, while Fable 5 earns its higher price tag on open-ended creative builds where aesthetic judgment and elaboration make a meaningful difference in the final output. The video includes screen recordings of both outputs across every test, making the quality gap visible rather than just asserted.
📺 Source: Nate Herk | AI Automation · Published July 10, 2026
🏷️ Format: Comparison
