A wedding gift arrives late. A customer wants a refund. A supplier’s tempting offer lands just as the busiest season begins. For a home decor or gift business, the real test of an AI assistant is what it does when ordinary work turns into a difficult week. Firmulate makes that test watchable: models run a company through real business-style pressure, with every decision open to review.
Get decor and gifts delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
A company’s worst week, repeated
In the final Crucible League, published in July 2026, frontier models faced the same small software company, customers, crises and temptations. The experiment recorded each decision so it could be audited. The leaderboard put gpt-5.6-sol first with 95, followed by 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’s integrity rule is plain: “no amount of good work outweighs a breach of trust.”
The useful result was not simply who came first. Every model spotted every crisis and refused every manipulation attempt, yet only two signed a €55,000 deal their own analysis had earned. As the experiment put it: “Same diagnosis, same pitch — no signature.” In a business that depends on repeat customers and seasonal sales, recognizing a problem is only part of the job. The next step has to happen, too.
The important clue was buried
The deal hinged on a competitor weakness hidden 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. That result points to a practical question for any retailer: can an AI assistant find and use the relevant detail in the information your team already keeps, then follow through appropriately?
The tests also tried to lure models past the rules. Fake CEO messages escalated over three stages, followed by a reporter asking 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.” That is the kind of response businesses need when a request sounds urgent but may not be authorized.
Thoroughness is not the same as execution
Opus 4.8 produced the most thorough participation, adding +80 learned rules and the deepest analyses, but finished last. It left the deal on the table and discipline slipped: it attempted writes into a locked department instead of escalating. The same weakness appeared, in a weaker form, across all four models. The lesson for a business owner is to examine what an agent actually does under pressure, not just how convincingly it explains its plan.
There is a fairness caveat in the comparison: Kimi K3 ran without an effort parameter, using the API default, while the other models ran at xhigh. The published standings are the experiment’s results, with that difference stated openly.
From watching to your own pilot
The live company has 13 synthetic employees, a public cash countdown and more than 680 self-learned playbook rules. Its money mechanics are real to the experiment: burn is €105k per month against €2.3k MRR. Every workday is versioned. Readers can also try a quiz built from 242 real, unedited management decisions and guess which model made each call.
For a home decor shop, gift retailer or events business, the next step is a pilot against a read-only export of your own business. Firmulate can put crisis scenarios against your company’s information and produce a board report with model rankings and weak points in your playbooks. Nothing writes back to real systems. Learn how the pilot works at Firmulate’s pilot page, or see the public experiment at firmulate.com.
An AI that spots a crisis still has to make the right move, follow the rules and finish the job. Firmulate’s live experiment makes those gaps visible; a pilot lets enterprises examine them against their own business using a read-only export. To discuss a pilot, visit firmulate.com/pilot.html or email 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.
