GPT 5.6-Sol vs. Claude Fable: Why OpenAI’s new model crushes my benchmark
Claire Vo compares OpenAI's new GPT 5.6 models (Soul, Terra, Luna) against Claude's Fable using her custom "How I AI" benchmark, finding that GPT 5.6 Soul excels at practical product work, prototyping, and natural communication, while Fable is theoretically intelligent but pedantic and difficult to collaborate with.
Summary
Claire Vo presents a comprehensive evaluation of OpenAI's newly released GPT 5.6 model family compared to Anthropic's Claude Fable. The three new GPT 5.6 variants include Soul (frontier/highest capability), Terra (balanced/efficient), and Luna (affordable/mini model). Pricing is significantly cheaper than Fable ($5 per million input tokens for Soul vs. $10 for Fable), with subscription access included in OpenAI's offering.
Vo developed a custom "How I AI" benchmark that evaluates models across PRD writing, prototyping, code debugging, and agentic voice (natural communication). The evaluation uses both an LLM-based judge (GPT 5.5) and Vo's own subjective "Clairvo taste test" scoring, weighted 70% human judgment and 30% machine judgment. Testing covered six models: Fable 5, Claude Sonnet 5, and the three GPT 5.6 variants.
Key findings show GPT 5.6 Soul ranked highest overall with the best taste scores across design prototypes and functionality. Vo praised Soul for creating unique, non-generic designs with strong visual hierarchy and semantic color usage. For specific tasks: Soul dominated prototype design, Terra was preferred for clean PRD writing, Sonnet 5 performed best at bug hunting and agentic voice (sounding most human-like), and Fable produced serviceable work but with less unique design aesthetic.
Vo's core criticism of Fable is that it communicates like "an engineer that has never met a human before," using excessive jargon, M-dashes, and pedantic language that makes collaboration difficult. She describes Fable as "theoretically intelligent" but "practically ineffective," getting stuck in its own frameworks and unable to break constraints appropriately. In contrast, Soul is described as straightforward, clear, and willing to reconsider limitations to ship working products.
Practical use cases where Soul excels include: zero-to-one prototyping (demonstrated with a gamified homework tracking app for her children), browser automation via the @Chrome tool in Codex (successfully managing 500+ LinkedIn messages), video editing/social media clipping, and building complex product tools. Vo also notes Soul has a distinctive design tendency toward forest green colors, appearing to stem from system prompts emphasizing "woodland elegance."
The evaluation reveals design philosophy differences: Claude's aesthetic tends toward editorial design with beige backgrounds, burnt orange, and italic serif fonts (which Vo dislikes), while Soul produces more opinionated, varied designs. Vo emphasizes she values uniqueness, creativity, functionality, and non-AI-like writing—despising "slop" (generic AI output patterns) above all.
Key Insights
- GPT 5.6 Soul is practically effective at shipping products while Fable is theoretically intelligent but struggles with real-world constraints; Vo characterizes Fable as creating hardened, over-engineered architectures that break when other models try to use them, whereas Soul is willing to loosen constraints to achieve user goals.
- Fable's communication style is described as inscrutable and pedantic—like an engineer unfamiliar with human communication—making it difficult to collaborate with as an end user, whereas Soul writes straightforwardly and clearly, enabling better human-AI collaboration.
- Soul demonstrates a systematic design bias toward forest green colors (possibly from a system prompt mentioning 'woodland elegance'), which Vo notes as a recognizable 'tell' that will become frustrating with repeated use.
- Vo's custom benchmark weights her subjective human judgment at 70% and LLM-based judging at 30%, because she 'decided to strike a balance' and 'like my own taste better,' showing deliberate preference for human taste over machine evaluation metrics.
- Zero-to-one prototyping in Soul was robust and functional, including complex features like gamified habit tracking with voice interaction, focus timers, and parent admin panels, suggesting Soul can generate consumer-grade prototypes in single shots that other models struggle to match.
Topics
Transcript
[0:00] I have been very very very sad the last week because for the last week I have not had access to my true favorite top-of-the-line model GPT56. But guess what babes? It is back and I am here to walk you through GPT56 Soul, GPT56 Luna, GPT56 Terra. I'm going to tell you what are these models, how have I been using them, why are they my [0:31] heart's favorite, and is Fable better than all of them or not? I have been testing this model for a couple weeks. There was a few days there where we didn't have access and I found myself desperate to get this workhorse model back. Now, we're not just relying on my…
Full transcript available for MurmurCast members
Sign Up to AccessMore from How I AI
Context offloading is an underrated AI use case
The speaker highlights context offloading as an underrated AI use case, where AI serves as a safety net for routine cognitive tasks like email management and personal finances. Rather than adding new capabilities, AI reduces anxiety about missing important information or making mistakes by handling monitoring tasks, thereby freeing up mental bandwidth.
WTF is a harness?`
A harness is code that wraps around an AI agent to make it more effective. The speaker demonstrates how a harness can automatically investigate production issues by analyzing Sentry warnings, identifying root causes, determining impact, and recommending whether to create tickets or apply fixes.
Stop prompting your AI agents. Start managing them.
The speaker discusses the shift from traditional agent prompting to agent management, highlighting the limitations of local Kanban board approaches and advocating for cloud-based VPS solutions with multiple communication channels to effectively manage autonomous AI agents across projects.
How to build a custom AI harness with Claude SDK
A custom AI harness is code wrapped around an AI agent to make it more effective for specific workflows. The speaker demonstrates building a Sentry bug-debugging harness using Claude SDK with a terminal UI, showing how structured constraints, custom prompts, and specific tool adapters enable agents to handle complex tasks more efficiently than general-purpose AI tools.
How I run autonomous coding agents from my phone with OpenAI Symphony + Linear
Alessio Finelli demonstrates how he uses OpenAI's Symphony framework combined with Linear and Codex to run autonomous coding agents from his phone, managing both software engineering tasks and a Pokémon card trading business through cloud-based VPS infrastructure instead of local machines.