Summary
Nate B Jones puts Claude Fable 5.1 through two demanding real-world knowledge-work tasks: generating a 37-second architectural walkthrough film using Blender from a single Seattle property address, and building a discounted cash flow model plus executive deck for GoPro’s acquisition by Starman. The results are compared across Fable 5.1 low and extra effort settings, against Fable 5 and GPT-5.6 Soul.
The financial modeling comparison is detailed. Fable 5.1 on low produced a seven-sheet Excel workbook and 13-slide deck with working formulas, scenario analysis, and a $1.15/share base case — all from a concise plain-language prompt. Fable 5.1 on extra delivered more thorough uncertainty analysis and polished design. GPT Soul stood out by adding a dedicated sources sheet and explicit formula check sheet, making the analysis far easier for a human analyst to audit — even though Soul’s visual design was less refined than Anthropic’s output.
The central argument distinguishes knowledge work from code: unlike software that passes or fails tests, deliverables like decks and spreadsheets require human judgment to evaluate completeness, which makes them a structurally harder class of AI task. Jones proposes treating model settings as sequential passes — low for a strong first draft, extra for depth and polish, Soul for auditability — rather than searching for a single best model. The 37-second Blender film, generated entirely from code without the creator knowing how Blender works, serves as the video’s most striking demonstration of Fable 5.1’s agentic capability.
📺 Source: AI News & Strategy Daily | Nate B Jones · Published September 04, 2026
🏷️ Format: Review
