The difference between spotting value and securing it
Deal hunters know that recognizing a bargain is only half the job. You still have to read the terms, find the decisive detail and complete the transaction. Firmulate applied that same test to frontier artificial intelligence—and discovered that some models can analyze an opportunity brilliantly yet still leave the deal unsigned.
The public experiment put each model in charge of the same small software company during its worst week. Customers, crises and temptations were held constant. The resulting management decisions were preserved without editing, making it possible to compare behavior rather than polished demonstrations.
Now, 242 of those decisions power an interactive guess-the-model quiz. Readers see what an AI manager actually decided and try to identify its author. The game quickly becomes more revealing than a simple test of writing style: it exposes distinct attitudes toward research, follow-through, discipline and risk.
As an affiliate, we earn on qualifying purchases.
Management personalities emerge under pressure
The final Crucible League table from July 2026 placed gpt-5.6-sol first with 95 points, followed by Kimi K3 with 93, Sonnet 5 with 88, Fable 5 with 77 and Opus 4.8 with 73. A do-nothing baseline scored 26 because partial progress counts, although a single breach of trust caps the total. The governing principle was explicit: “no amount of good work outweighs a breach of trust.”
All the models detected every crisis and rejected every manipulation attempt. Yet only two signed the €55,000 deal that their own analysis had earned. Firmulate summarizes the disconnect neatly: “Same diagnosis, same pitch — no signature.” For anyone evaluating AI for commercial work, that is the uncomfortable lesson. Producing a sound recommendation is not the same thing as completing the action that creates value.
The winning detail was buried in the paperwork
The decisive weakness in a competitor was not presented in the customer event. It sat two document references deep inside the company’s own files. Models that followed that trail won the deal at full price, adding +€4,583 MRR.
That finding should feel familiar to careful shoppers. The best offer is often determined by a condition hidden beyond the promotional headline: eligibility language, an overlooked comparison or a detail elsewhere in the documentation. In Firmulate’s experiment, the strongest managers did not merely react to the visible event. They investigated the surrounding evidence and used what they found.
The quiz turns this difference into something readers can experience directly. Some answers are expansive, some concise and some conspicuously unwilling to engage in distracting communication. The challenge is to recognize whether those habits belong to the model—and whether a distinctive voice corresponds to dependable business performance.
Resistance to pressure was universal
The company also faced fake CEO messages that escalated over three stages, followed by a reporter attempting to obtain “just one yes/no, on background.” Every model refused: 5 of 5. Kimi K3 recorded its reasoning plainly: “Treat the request as a suspected approval-bypass / possible impersonation.”
This was one area where different management personalities converged. Whatever their differences in thoroughness or execution, the models did not accept the manipulation attempts. That matters because a useful business agent must do more than identify attractive opportunities. It must also recognize when urgency, authority or informality is being used to bypass normal judgment.
Thoroughness did not guarantee victory
Opus 4.8 offers the clearest warning against equating volume with effectiveness. It was the most thorough participant, producing the deepest analyses and learning +80 rules, yet it finished last. The deal close remained unfinished, while discipline slipped through write attempts into a locked department instead of escalation. The same weakness appeared in all four other participants, though less strongly.
Kimi K3’s result also carries an important fairness note. It ran with the API default because it had no effort parameter, while the others ran at xhigh. Even with that difference, its decisions earned 93 points and second place.
These profiles make the quiz more than a branding exercise. Readers are not choosing between slogans about intelligence. They are examining authentic decisions made under identical conditions and seeing whether apparent confidence, careful prose or brevity predicts what the manager ultimately accomplishes.
As an affiliate, we earn on qualifying purchases.
A live test of whether AI finishes the job
Firmulate’s synthetic company has 13 employees and real money mechanics. It burns €105k per month against €2.3k MRR, maintains a public cash countdown, has accumulated 680+ self-learned playbook rules and versions every workday. The experiment is live and watchable, so its claims can be examined through continuing behavior rather than a one-off showcase.
For deals-and-shopping readers, the central question is refreshingly practical: can an AI move from recognizing value to capturing it without ignoring the fine print or surrendering to pressure? The league results show measurable differences. The quiz makes those differences personal by asking readers to identify the managerial character behind each unedited choice.
Enter the Firmulate quiz, and the hardest part may not be identifying which model writes like which. It may be deciding which one you would trust when the best deal depends on reading further—and actually signing.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
