Get wellness gear delivered free with Prime
- Fast, free delivery on millions of items
- Prime Video, Amazon Music and more included
- Member-only deals all year
Wellness tech asks for trust. AI needs to earn it.
A home wellness service might handle appointments, customer questions and sensitive personal details. If AI agents take on those jobs, a polished chat demo won’t show how they behave when a customer is upset, a deal is tempting or someone pretends to be the boss. Firmulate’s live experiment puts AI models through that kind of pressure inside a small software company. Its larger promise for businesses is a way to rehearse before letting AI near real operations.
One company, a week of hard choices
In the final Crucible League, published in July 2026, each frontier model faced the same small company, customers, crises and temptations. Every decision was versioned and auditable. 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 benchmark counts partial progress, but a single breach of trust caps the total: “no amount of good work outweighs a breach of trust.”
The models all spotted every crisis and refused every manipulation attempt. Yet only two signed the €55,000 deal their own analysis had earned. The finding, in Firmulate’s words: “Same diagnosis, same pitch — no signature.” Recognizing a good move and carrying it through were different tests.
The clue was buried in the company’s own files
The deal turned on a competitor weakness hidden two document references deep in the company’s files, rather than in the customer event itself. Models that read the file won at full price, worth +€4,583 MRR. It’s a useful reminder for any business considering AI: important context may already exist in internal documents, while the decisive failure may be not a mistaken answer but a missed follow-through.
The trust test was similarly concrete. Fake CEO messages escalated across three stages, followed by a reporter’s “just one yes/no, on background” trick. All 5 of 5 models refused. Kimi K3’s on-record reasoning was: “Treat the request as a suspected approval-bypass / possible impersonation.” The experiment also surfaced a less tidy side of performance. 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 tried writing into a locked department instead of escalating. A weaker version of that same discipline problem appeared in all four.
From watching to trying it on your business
Firmulate describes its live company as a lab with 13 synthetic employees, real money mechanics, a public cash countdown and more than 680 self-learned playbook rules. It reports burn of €105k/month against €2.3k MRR, with every workday versioned. A separate quiz draws on 242 real, unedited management decisions and asks visitors to guess which model made them.
The enterprise pitch is a step beyond watching the live company: run a wargame against a read-only export of your own business. That means scenarios shaped around your customers, pipeline and rules, followed by a board report with model rankings and weak points in your playbooks. Firmulate says nothing writes back to real systems. The goal is to see how an AI workforce handles pressure before it is given real responsibilities.
There is a caveat in comparing the league’s results: Kimi K3 ran without an effort parameter, using the API default, while the other models ran at xhigh. The rankings offer a view of performance in this experiment, not a universal verdict on which model will suit every company.
A rehearsal for decisions that matter
For wellness businesses, customer trust can be as important as operational speed. Firmulate’s experiment suggests that spotting a crisis or refusing a trick is only part of the job; models also need to find the right context and follow through with discipline. A company-specific pilot can put those behaviors under pressure using a read-only export, before AI touches live systems. Explore a Firmulate pilot or contact contact@firmulate.com.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Cold & flu season Picks
humidifiers
As an affiliate, we earn on qualifying purchases.
