GPT-5.6 Soul outshines Fable 5 for practical GTM AI tasks

Jul 9, 2026 · Lenny's Podcast
🎧 PodShort 37 min squeezed to 3 AI SprinklerAS AI / ML
Episode artwork
Clare
Host, How I AI Podcast; Product Leader at How I AI Podcast; Chat PRD
Lenny's Podcast
37 min squeezed to 3
Full episode from Lenny's Podcast
Quotable Moments

I hate talking to Fable 5 because it talks to me like an engineer that has never met a human before. It's like its first day on Earth.

ClareHost, How I AI Podcast; Product Leader, Chat PRD
08:29

Fable is theoretically hyper intelligent and Soul is practically effective.

ClareHost, How I AI Podcast; Product Leader, Chat PRD
21:23

Please, please, please learn to use @Chrome, @Browser, and @Computer and just let Codex rip and let GPT 5.6 rip.

ClareHost, How I AI Podcast; Product Leader, Chat PRD
34:56
Key Insights
  • GPT-5.6 Soul is the host's preferred AI model, based on a 70/30 split between her personal 'taste test' and an LLM judge, due to its practical effectiveness and superior outputs across various tasks.
  • The 'How I AI benchmark' rigorously evaluates AI models on critical tasks including generating good Product Requirement Documents (PRDs), wireframing, creating robust prototypes, debugging code, and engaging in human-like communication.
  • Fable 5 is characterized as 'theoretically hyper intelligent' but 'practically ineffective' due to its pedantic communication style and rigid, self-limiting frameworks, which significantly hinders user collaboration and flexibility.
  • GPT-5.6 Soul consistently delivers unique, functional, and aesthetically pleasing designs, especially in complex front-end prototyping, making it stand out from competitors that often produce generic layouts.
  • GPT-5.6 Soul is significantly more affordable than Fable 5, with input tokens costing $5 per million and output tokens $30 per million, compared to Fable's $10 and $50 respectively.
  • GPT-5.6 demonstrates exceptional capabilities in browser automation, effectively managing tasks like bulk LinkedIn messaging, filling out forms, and testing web applications with high efficiency.
  • The model can rapidly generate high-quality, gamified homework tracking systems for children in a single shot, showcasing its ability to build robust and opinionated zero-to-one prototypes with detailed functionality and engaging design.
  • GPT-5.6 Soul is highly proficient in video editing tasks, capable of quickly clipping and producing 'hype videos' from longer recordings, with potential for color grading and sound adjustments.
Metrics Mentioned
  • 50 written reactions (Clare's qualitative evaluation input for the models, where some were 'unmistakable hits' and others 'garbage'.)
  • 70% Clare's taste / 30% LLM judge split (The weighted index used to determine the final ranking and preference for AI models.)
  • $5 per million input tokens (The pricing for GPT-5.6 Soul.)
  • $30 per million output tokens (The pricing for GPT-5.6 Soul.)
  • $10 per million input tokens (The pricing for Fable 5, twice as expensive as GPT-5.6 Soul.)
  • $50 per million output tokens (The pricing for Fable 5, significantly more expensive than GPT-5.6 Soul.)
  • 500 messages (The approximate number of LinkedIn messages GPT-5.6 processed and responded to for browser automation.)

RevBots.ai View:

  • AI Sprinkler teams will chase GPT-5.6 for cost savings on content and automation.
  • Practical effectiveness beats theoretical precision for ARM-stage adoption.
  • Browser automation use cases (LinkedIn, forms) appeal to SaaS Hoarders.
  • Video editing capabilities could accelerate ARM-stage content ops.
🎧Full Episode:Lenny's Podcast →