GPT 5.6 Soul vs Fable 5: Real-World Showdown
After hands-on testing of GPT 5.6 Soul and Fable 5 across creative coding, web design, and API tasks, Soul dominates on cost and speed (roughly half the price, 24 wins vs 3 in quick tasks) but Fable remains superior for complex reasoning and creative work. Soul is a high-efficiency executor; Fable is a strategic manager. Choose based on your use case.
Browser Game Challenge
Game Quality: Fable Wins Despite Similar Build Time
Both models built playable browser bike games in roughly the same time (Fable 23:37, Soul 23:00), but Fable's 3D GTA-style open-world output was significantly more polished and immersive than Soul's top-down bird's-eye view. Fable's superior creative execution came at a cost premium.
Cost Gap: Soul 3x Cheaper for Same Task
Soul's game cost $4.50 with 31,000 output tokens, while Fable's cost $14.22 with 90,000 output tokens. Despite similar build times, Fable consumed nearly 3x the tokens, demonstrating Soul's token efficiency even when output quality was lower.
Interactive Website Challenge
Fable's Immersive Design Beats Soul's Efficiency
Fable created a 10-billion-year cosmic journey website with superior 3D interactivity and wow factor, while Soul produced a functional but less immersive 'Vesper Archive' experience. Fable's creative vision was on a different level despite Soul's speed advantage.
Cost Disparity Widens: Fable 20x More Expensive
Fable's website cost $19.24 with 80,000 output tokens and took 23 minutes, while Soul's cost $1.01 with 20,000 tokens in under 7 minutes. The 20x cost difference raises the question: was Fable's output 20x better? The tester concluded it was not.
Five Creative Elements Challenge
Soul Edges Out Fable on Diversity and Concept
When asked to design five fundamentally different visual elements (games, presentations, websites, simulations), Soul produced more cohesive and diverse outputs with better aesthetic consistency, while Fable's outputs felt less unified. Soul won this round on creative direction despite both having bugs.
Speed Surprise: Fastest Task Despite Complexity
Counterintuitively, this most complex prompt (five different things) was the quickest: Fable took 15 minutes and Soul took 7 minutes. Soul was nearly 2x faster while maintaining comparable creative quality, suggesting Soul's efficiency scales well with ambiguous, multi-part requests.
Cost Efficiency Consistent: Soul ~$1, Fable ~$15
Soul cost $1.22 with 22,000 output tokens; Fable cost $15 with 65,000 tokens. The pattern holds: Soul is roughly 12x cheaper while maintaining acceptable quality, confirming token efficiency as a core advantage.
Quick API One-Shot Tasks
Soul Dominates Quick Tasks: 24 Wins vs 3
In rapid-fire API calls (stateless, one-off requests), Soul won 24 times while Fable won only 3 times. Fable frequently refused to answer due to strict safety guardrails, whereas Soul handled requests more liberally. When Fable did answer, quality was comparable (Fable 0.966 vs Soul 0.98 score).
Cost Gap Narrows on API: Soul $9 vs Fable $14
For API-only calls, Soul cost $9 total while Fable cost $14—a $5 gap. This is much smaller than the agentic reasoning gap, suggesting the cost advantage is partly driven by multi-step reasoning loops rather than pure model pricing.
Token Efficiency: Soul Half the Cost of Fable
Across all tests, Soul's input and output tokens are roughly half of Fable's. This efficiency gap persists even when quality is comparable, suggesting Soul's training or architecture yields better token utilization independent of task complexity.
Latency: Soul More Variable, Fable More Consistent
Soul's median API latency was faster than Fable's, but Fable's mean (average) was faster. This indicates Soul has higher variance—some requests are very quick, others very slow—while Fable hovers consistently around 20 seconds. For predictable performance, Fable is more reliable.
Model Positioning & Use Cases
Fable: The Manager, Soul: The Worker
Fable excels at orchestration, reasoning, creativity, and strategic advising—best suited for complex decision-making and high-stakes work. Soul is a high-efficiency executor: fast, cheap, and reliable for shipping features and handling verification tasks. Ideally, Fable orchestrates Soul agents.
Fable's Strengths: Creativity, Writing, Video Production
Fable is superior at brainstorming, creative writing, strategic advising, and video production. When the tester had each model create YouTube videos from scratch, Fable's output was noticeably better. Fable avoids overengineering and delivers polished, intentional results.
Soul's Strengths: Speed, Cost, Computer Use, Devil's Advocate
Soul excels at rapid execution, computer use (vision/interaction), and critical analysis. It's more careful about following instructions exactly and finding bugs from multiple angles. Soul's lower cost and speed make it ideal for high-volume, lower-complexity tasks and verification workflows.
Capability Gap: Soul vs Fable vs Opus
Soul sits clearly above GPT 5.5 and Opus 4.8 (which are very similar), but there remains a substantial gap between Soul and Fable 5. Fable is in another tier for pure capability. Soul is roughly equivalent to Opus 4.8 in power but cheaper. The tester speculates GPT 6 will be Fable's true competitor.
Pricing Reflects Capability: Soul Priced Like Opus, Not Fable
Soul's pricing aligns with Opus 4.8, not Fable 5, because Soul's capability is comparable to Opus. If Soul were truly Fable-tier, it would be priced at Fable's premium. The tester notes Soul is labeled 5.6 (not 6.0), suggesting Anthropic is holding back for a future release.
Key Takeaways
Benchmarks Don't Tell the Full Story
While benchmarks show Soul outperforming Opus and Fable, real-world testing reveals Fable is still superior for complex creative and strategic work. Benchmarks measure narrow capabilities; actual use cases require hands-on evaluation of output quality, speed, and cost trade-offs.
Unit Economics Matter: Right Tool for the Task
If a task only needs a capability level of 5-6, using a level-10 model (Fable) is wasteful. Soul's efficiency makes it ideal for high-volume, routine work. Fable justifies its cost only for work requiring top-tier reasoning and creativity. Match model tier to task complexity.
Use Case Determines Winner, Not Benchmarks
Some users may prefer Soul; others Fable. The choice depends on your workflow, harness setup, instruction style, and priorities (cost vs quality vs speed). Hands-on testing with your actual work is the only reliable way to decide.
Notable quotes
Soul is just a really, really good worker. Fable is a better manager. — Nate Herk
You have to get your hands dirty. You have to do actual work that you do on your day-to-day. — Nate Herk
Fable 5 is in another tier. It's just on another level. — Nate Herk
Action items
- Run your own side-by-side tests with Soul and Fable on your actual day-to-day work tasks before committing to either model.
- Calculate your unit economics: if your task only needs capability level 5-6, use Soul to save 70-80% on costs; reserve Fable for high-complexity reasoning and creative work.
- Test Soul for high-volume API calls and verification workflows; use Fable for strategic planning, creative direction, and complex multi-step reasoning.
- Monitor token efficiency in your own usage; track whether Soul's lower token consumption translates to meaningful cost savings in your harness or workflow.
- Watch the two YouTube videos mentioned (Fable made this video / Soul made this video) to compare creative output quality firsthand.