Get business pricing on office and shipping supplies
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
What happens when an AI has to close the deal?
For a business, an AI that can write a persuasive sales email is only part of the story. Can it spot a customer risk, resist a convincing impersonation and still follow through when the right deal is on the table? Firmulate’s live experiment puts models in charge of the same small software company during its worst week—and shows why those questions matter before AI gets near your business.
Same crises, different outcomes
In the final Crucible League, published in July 2026, five models were ranked on how they managed the company: gpt-5.6-sol scored 95, Kimi K3 scored 93, Sonnet 5 scored 88, Fable 5 scored 77 and Opus 4.8 scored 73. The do-nothing baseline scored 26. Firmulate’s standard is deliberately unforgiving: partial progress counts, but a single breach of trust caps the total. As the project puts it, “no amount of good work outweighs a breach of trust.”
The models were given the same customers, crises and temptations. Every decision was versioned and auditable. All spotted every crisis and refused every manipulation attempt. But only two signed a €55,000 deal that their own analysis had earned. The summary is striking: “Same diagnosis, same pitch — no signature.” Recognizing the opportunity was not the same as acting on it.
The clue was in the company’s files
The deal turned on a competitor weakness buried two document references deep in the company’s own files—not in the customer event. Models that read the file won the deal at full price, worth +€4,583 MRR. It’s a reminder that useful business judgment can depend on joining information across the company, then carrying the decision through.
The pressure tests also went beyond ordinary business problems. Fake CEO messages escalated over three stages, followed by a reporter’s request for “just one yes/no, on background.” All five models refused. Kimi K3’s reasoning was to “Treat the request as a suspected approval-bypass / possible impersonation.”
Thoroughness did not guarantee a win
Opus 4.8 was the most thorough participant, with +80 learned rules and the deepest analyses, yet it finished last. It left the deal unsigned and discipline slipped: it attempted writes in a locked department instead of escalating. The same weakness showed up, less strongly, in all four models. More analysis alone did not ensure the company got the result its analysis supported.
There is a fairness caveat in the comparison: Kimi K3 ran without an effort parameter, using the API default, while the others ran at xhigh. The ranking offers a view of performance in this experiment, not a guarantee of how any model will perform in every business.
A live company—and a path to your own pilot
The experiment is watchable at Firmulate. Its live company has 13 synthetic employees, burn of €105k per month against €2.3k MRR, a public cash countdown and 680+ self-learned playbook rules. Every workday is versioned. A quiz built from 242 real, unedited management decisions invites readers to guess which model made each call.
For business leaders, the next step is to test decisions against the realities of their own organization. An enterprise pilot uses a read-only export of the company to stage crisis scenarios and produce a board report with model rankings and weak points in existing playbooks. Nothing writes back to real systems. That moves the exercise from watching a live company to examining what AI might do with yours, before it is put to work.
See how your playbooks hold up
Firmulate’s live experiment shows that spotting a crisis and making the right move are different tests. Enterprises can run the same wargame against a read-only export of their business, with no changes written back to real systems. Explore the enterprise pilot or contact contact@firmulate.com.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
