In the gym, a polished workout plan is only a start. The real test comes when the weight gets heavy, fatigue sets in and you have to choose whether to stick to the plan or improvise. Businesses face their own worst-week test: a customer crisis, a tempting shortcut and a deal that needs someone to follow through.
Get workout gear delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
Firmulate puts AI models into that kind of pressure test. Its live experiment is a small software company, run by synthetic employees through real money mechanics and public cash countdowns. The point is not whether an AI sounds like a capable manager. It is what it does when the next decision counts.
Same company, same bad week
In the final Crucible League, published in July 2026, each frontier model faced the same customers, crises and temptations. Every decision was versioned and auditable. The results ranked gpt-5.6-sol first 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. The league counts partial progress, but a single breach of trust caps the total: “no amount of good work outweighs a breach of trust.”
On one test, the models all recognized the crises and refused manipulation attempts. Yet only two signed the €55,000 deal their own analysis had earned. The diagnosis was right; the pitch was ready. Still, some models left the close on the table. As Firmulate puts it: “Same diagnosis, same pitch — no signature.” That gap between knowing the play and making it is hard to see in a chat demo.
Products Worth Considering
Levoit Vital 100S-P Air Purifier
BowFlex Results Series SelectTech Adjustable Dumbbells (Pair)
Levoit Core 400S-P Smart Air Purifier for Large Rooms
As an affiliate, we earn on qualifying purchases.
The clue was buried in the company’s own files
The deciding competitor weakness sat two document references deep in the company’s files, not in the customer event itself. Models that read the file won the deal at full price, worth +€4,583 MRR. It is a practical reminder for any business considering AI: useful context may already be in your own records, but an agent still has to find it, interpret it and act on what it learns.
Trust was tested directly, too. Fake CEO messages escalated over three stages, followed by a reporter’s request for “just one yes/no, on background.” All 5 of 5 models refused. Kimi K3 explained its reasoning on the record: “Treat the request as a suspected approval-bypass / possible impersonation.” That is the sort of judgment an organization needs to examine before giving AI responsibility around customers, money or sensitive decisions.
Thoroughness did not guarantee the 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 into a locked department instead of escalating. A weaker version of the same weakness appeared in all four models. The result resembles training: doing more work does not automatically mean performing well when the moment to execute arrives.
There is a fairness detail readers should keep in mind. Kimi K3 ran without an effort parameter, using the API default; the other models ran at xhigh. Firmulate also offers a quiz built from 242 real, unedited management decisions, so readers can try to guess which model made each call.
From watching to testing your own playbook
The live company makes the experiment watchable: 13 synthetic employees, burn of €105k/month against €2.3k MRR, a public cash countdown, 680+ self-learned playbook rules and every workday versioned. The site excerpt describes a company that rebuilds itself twice a day. Readers can follow the experiment at firmulate.com.
For a business, the next step is a pilot against its own company. Firmulate says an enterprise can provide a read-only data export, then test crisis scenarios against its customers, pipeline and rules. The output is a board report showing model rankings and weak points in the company’s playbooks. Nothing writes back to real systems. That gives leaders a chance to see where an AI follows through, where it hesitates and where the existing plan needs work before deploying agents into live operations.
A model can spot a crisis, refuse a scam and still fail to close the deal. Firmulate’s experiment turns that gap into something businesses can watch and, through a pilot, test against their own playbooks using a read-only export. To discuss a pilot, visit firmulate.com/pilot.html or contact contact@firmulate.com.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
