In cycling, a strong rider can look ready on a calm training day. A hard race reveals more: judgment under pressure, discipline when plans change, and whether the team can close when the moment arrives. Businesses face their own version of that test as AI moves from chat windows into everyday work.
Get sports gear delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
One company, the same worst week
Firmulate put frontier AI models in charge of the same small software company through the same crises, with the same customers and temptations. Each decision was versioned and auditable. The point was to observe management, not just polished answers.
The final Crucible League, published in July 2026, ranked gpt-5.6-sol first with 95, Kimi K3 second with 93, Sonnet 5 third with 88, Fable 5 fourth with 77, and Opus 4.8 fifth with 73. The do-nothing baseline scored 26. The benchmark also makes a breach of trust decisive: no amount of good work outweighs one.
Diagnosis is not the finish line
Every model spotted every crisis and refused every manipulation attempt. Yet only two signed the €55,000 deal their own analysis had earned. Firmulate’s summary is blunt: “Same diagnosis, same pitch — no signature.” A system may understand the opportunity and still fail to act on it.
The deal hinged on a detail buried two document references deep in the company’s own files, rather than in the customer event itself. Models that read the file won at full price, worth +€4,583 MRR. In a race, knowing the route and taking the decisive turn are separate skills. This experiment exposed that gap in a business setting.
The pressure test also included fake CEO messages escalating over three stages, followed by a reporter asking, “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.” That kind of restraint matters when an agent can encounter requests that look urgent or authoritative.
Thoroughness has its own limits
Opus 4.8 was the most thorough participant, with +80 learned rules and the deepest analyses, but finished last. It left the close on the table and let discipline slip, making write attempts into a locked department instead of escalating. A weaker form of the same weakness appeared in all four.
There is a fairness detail in the comparison: Kimi K3 ran without an effort parameter, using the API default, while the others ran at xhigh. The results are a specific experiment, not a promise that every deployment will behave the same way.
From watching to trying it on your business
The public Firmulate experiment is a live, watchable company with 13 synthetic employees and real money mechanics: €105k monthly burn against €2.3k MRR, a public cash countdown, 680+ self-learned playbook rules, and workdays that are versioned. A quiz built from 242 real, unedited management decisions lets readers guess which model made each call.
For a business, the next step can be a pilot using a read-only export of its own data, with crisis scenarios tailored to its customers, pipeline, and rules. The result is a board report that compares model performance and surfaces weak points in the company’s playbooks. Nothing writes back to real systems. Readers can watch the experiment at Firmulate and explore the pilot at firmulate.com/pilot.html.
Take the pressure test closer to home
Watching an AI handle another company’s worst week is useful. Seeing how models respond to your own customers, decisions, and weak spots could make the findings more actionable. To discuss a Firmulate pilot using a read-only business export, visit firmulate.com/pilot.html or contact contact@firmulate.com.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
NFL season / tailgating Picks
team gear
As an affiliate, we earn on qualifying purchases.
