Imagine trusting a fashion stylist to only pick the trendiest outfits — but instead, they ignore the latest runway shows. When it comes to AI in business, trust is equally critical. Just like in style, performance isn’t enough; integrity and reliability matter more. That’s what a recent public experiment with AI models shows — even the simplest baseline scores 26 points, setting a meaningful floor in measuring true performance.
Get your wardrobe favorites delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
In the fast-paced world of fashion, we often assume that more elaborate choices mean better style. But in the realm of AI for business management, superficial metrics can be misleading. That’s why the recent Firmulate experiment is illuminating: it pits top-tier AI models against a simple, do-nothing baseline to understand what real trust and performance look like in managing a business’s toughest week.
The Benchmarking Methodology: Simulating Business Under Pressure
Four frontier AI models were tasked with running the same small software company through its worst week — with identical customers, crises, and temptations. Every decision was recorded and auditable, ensuring transparency. The goal was to see if these models could handle real-world scenarios where honesty and strategic judgment matter as much as technical prowess.
Interestingly, even the do-nothing baseline, which essentially does nothing but passively exists, scored 26 points. This score isn’t a typo; it reflects that partial progress counts. If the system reads the files and recognizes a crisis, that counts toward the score, even if it doesn’t act perfectly. More importantly, a single breach of trust — like signing a deal the analysis advises against — caps the total score at 26, no matter how competent the model appears otherwise.
Why Does a Do-Nothing Model Score 26?
This baseline score underscores a key principle: in measuring AI performance, honesty and adherence to ethical boundaries are non-negotiable. The experiment shows that even the simplest system, which does little beyond reading information, begins with a floor score. It indicates that trustworthy behavior is a critical component, and models that violate that trust cannot score higher, regardless of their other capabilities.
How Do Top Models Perform?
At the top of the league stands GPT-5.6-sol with an impressive score of 95, having identified the buried fact that led to closing a lucrative deal. Kimi K3 follows with 93 points, demonstrating the cleanest discipline — it also closed the deal, just like GPT-5.6-sol. Sonnet 5 scored 88, and Fable 5 brought up the rear with 77, each closing deals but with some slips in process discipline.
It’s notable that only two models managed to sign the €55,000 deal their own analysis earned. All four models spotted every crisis and refused manipulative tactics, like fake CEO messages or reporter tricks. They showed integrity in difficult moments, which is crucial in real business decisions.
The Hidden Weaknesses: Reading Files Matters
The experiment uncovered a subtle flaw: models that read and analyze company files deeply had a decisive edge in closing deals at full price — worth over €4,583 in monthly recurring revenue (MRR). This reveals that effective comprehension, not just surface-level responses, can be the difference between a good decision and a missed opportunity.
Trust in Practice: Managing Under Pressure
All models refused social engineering attempts, including staged fake messages from a CEO and a reporter’s background request. As Kimi K3 explained, they treated such requests as potential impersonation or approval bypass, highlighting an inherent trustworthiness that’s vital in sensitive business environments.
The Live Experiment: Real Money, Real Crises
The entire test occurred within a live, simulated company environment with 13 synthetic employees, real revenue mechanics, and a public cash countdown. The company burns €105k monthly against €2.3k in MRR, emphasizing the importance of smart decision-making. This setup, which can be watched at firmulate.com/live, brings transparency to the process, allowing anyone to see how AI models handle crises in real time.
What Does This Mean for Business?
In the fashion world, we know that trust and authenticity are everything. The same applies to AI in business. The experiment from Firmulate shows that AI models capable of honest, disciplined decisions perform better — and that superficial metrics or chat demos can mask underlying weaknesses. A top AI isn’t just one that talks well, but one that acts reliably, reads deeply, and stays honest under pressure.
The Final Word: Setting a Trustworthy Standard
The benchmark’s lowest score of 26 points reflects an honest baseline of performance — one that values trust and integrity above all. As AI continues to play an increasing role in managing business operations, understanding these real-world capabilities will be vital. Just like selecting the perfect outfit, choosing a trustworthy AI model isn’t about flash, but about authenticity and reliability.
To explore the full results or try your hand at the quiz that tests management decisions against AI models, visit firmulate.com/benchmarks.html. When it comes to AI for business, performance is important, but integrity is everything.
Trustworthiness in AI isn’t just a bonus — it’s a baseline. A simple do-nothing model scores 26, highlighting that honest, disciplined decision-making is the true measure of AI readiness for business crises.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
As an affiliate, we earn on qualifying purchases.
Evergreen bestsellers Picks
bestsellers
As an affiliate, we earn on qualifying purchases.
