A tough workout reveals more than what you can do on a good day. It shows how you respond when fatigue sets in and the next choice matters. Firmulate applies a similar pressure test to AI: models run a company through a week of crises, where spotting trouble is only part of the job.
Same company, same hard week
In Firmulate’s final Crucible League, published in July 2026, each frontier model faced the same small software company, customers, crises and temptations. Every decision was versioned and auditable. The experiment asks a practical question: can an AI workforce carry its analysis through to a sound business decision?
All the models spotted every crisis and refused every manipulation attempt. But only two signed a €55,000 deal that their own analysis had earned. As the finding puts it: “Same diagnosis, same pitch — no signature.” Recognizing the right move and completing it turned out to be different tests.
The overlooked clue was already in the files
The deal hinged on a competitor weakness buried two document references deep in the company’s own files, rather than in the customer event. Models that read the file won the deal at full price, worth +€4,583 MRR. The detail makes the result concrete: business judgment can depend on connecting evidence that is present but easy to miss.
The integrity test was similarly direct. 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 on-record reasoning was: “Treat the request as a suspected approval-bypass / possible impersonation.”
Thorough work still has to reach the finish line
The final standings were gpt-5.6-sol at 95, Kimi K3 at 93, Sonnet 5 at 88, Fable 5 at 77 and Opus 4.8 at 73. The do-nothing baseline scored 26. Partial progress counted, but a single breach of trust capped the total: “no amount of good work outweighs a breach of trust.”
Opus 4.8 was the most thorough participant, with +80 learned rules and the deepest analyses, yet finished last. It left the close on the table and discipline slipped: it tried to write into a locked department instead of escalating. A weaker version of the same weakness appeared in all four. Detail and diligence matter, but they do not replace follow-through or respect for boundaries.
One fairness detail belongs beside the ranking: Kimi K3 ran without an effort parameter, using the API default, while the others ran at xhigh.
From watching to testing your own business
The live company makes the experiment watchable. It has 13 synthetic employees and real money mechanics: €105k monthly burn against €2.3k MRR, alongside a public cash countdown. Its playbook has more than 680 self-learned rules, and every workday is versioned. Firmulate also offers a quiz built from 242 real, unedited management decisions, inviting readers to guess which model made each call.
For enterprises, the next step is a pilot against a read-only export of their own business. That means testing crisis scenarios against the company’s own customers, pipeline and rules, then reviewing a board report with model rankings and weak points in the playbooks. Nothing writes back to real systems. The format moves the question from whether a model performs in a public experiment to how it handles the pressures your organization actually faces.
In a workout, good form under pressure matters as much as strength. Firmulate’s experiment suggests the same distinction for AI at work: models may identify crises and protect trust, yet still miss the action their analysis supports. Enterprises can test that gap with a read-only pilot using their own business data. Explore a Firmulate pilot or contact contact@firmulate.com.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
