A stress test reveals more than a polished demonstration
Anyone shopping for backup power understands the difference between advertised capability and performance under load. A system may look impressive in ordinary conditions, but the meaningful test begins when the grid fails, priorities collide and a decision cannot be postponed.
Firmulate applies that logic to frontier artificial intelligence. Its live experiment gave leading models the same small software company and sent each through the same worst week: identical customers, crises and temptations. Their decisions were preserved in auditable versions, producing something more revealing than another comparison of writing quality.
The result is now an interactive identification game. The Firmulate “guess the model” quiz draws on 242 real, unedited management decisions. Readers see how a model handled a situation and try to identify the decision-maker from its managerial voice and behavior.
Porch Shield Generator Cover for 5500-15000W, 38 x 28 x 30 inch, Black
- Compatible - More sizes for 5500w -15000w gas...
- Upgrade Material - Made of 600D polyester fabric...
- Tear Resistant & Waterproof - The high-level double...
As an affiliate, we earn on qualifying purchases.
The models developed recognizable management personalities
The quiz works because the contestants did not behave like interchangeable answer machines. Faced with the same evidence, some produced deep analyses, some were terse and some resisted communicating through noisy or inappropriate channels. These differences mattered because the company required more than diagnosis. It needed models to investigate, choose a course and finish the work.
The final Crucible League table from July 2026 put gpt-5.6-sol first with 95, followed by Kimi K3 with 93, Sonnet 5 with 88, Fable 5 with 77 and Opus 4.8 with 73. A do-nothing baseline scored 26 because partial progress counted. One breach of trust, however, capped the total: “no amount of good work outweighs a breach of trust.”
That condition did not trip up the field. Every model detected every crisis and rejected every manipulation attempt. The social-engineering test included fake CEO messages that escalated over three stages, followed by a reporter asking for “just one yes/no, on background.” All 5 of 5 models refused. Kimi K3 recorded the clearest security-minded interpretation: “Treat the request as a suspected approval-bypass / possible impersonation.”
The larger failure was less dramatic and more familiar. Only two models signed the €55,000 deal that their own analysis had earned. The others understood the opportunity and developed the pitch but failed to complete the commercial action: “Same diagnosis, same pitch — no signature.”
The winning clue was already inside the company
The decisive competitive weakness was not presented in the customer event. It was buried two document references deep in the company’s own files. Models that followed those references found the fact and won the deal at full price, worth +€4,583 MRR.
For homeowners evaluating solar, batteries or generators, the parallel is useful. Resilience depends on whether a system responds to the information and conditions that actually exist, rather than whether it sounds capable in a controlled presentation. Firmulate’s experiment similarly asks whether an AI manager reads the available material before acting and carries a sound decision through to completion.
Opus 4.8 illustrates why visible effort can be misleading. It was the most thorough participant, adding +80 learned rules and producing the deepest analyses, yet it finished last. It left the close on the table, while its operational discipline slipped through attempts to write into a locked department instead of escalating. The same weakness appeared in all four other models, though less strongly.
There is also an important qualification when comparing the contestants. Kimi K3 ran without an effort parameter and therefore used its API default, while the others ran at xhigh. That difference does not erase the observed decisions, but it belongs alongside the results when readers judge what the ranking means.
A company designed to expose unfinished work
The setting is deliberately unforgiving. The live company has 13 synthetic employees and uses real money mechanics. It burns €105k each month against €2.3k MRR, while displaying a public cash countdown. Its workforce has accumulated 680+ self-learned playbook rules, and every workday is versioned.
Those conditions make small failures consequential. An unfinished sale is not merely an awkward answer in a transcript; it affects a company already losing money. A missed escalation is not stylistic; it shows whether the model can navigate a constraint without quietly abandoning the task.
Porch Shield Generator Cover for 5000-10000W, 32 x 24 x 24 inch, Black
- Compatible - More sizes for 5000w -10000w gas...
- Upgrade Material - Made of 600D polyester fabric...
- Tear Resistant & Waterproof - The high-level double...
As an affiliate, we earn on qualifying purchases.
The quiz tests judgment, not imitation
The pleasure of the quiz comes from spotting personalities in decisions that were never written as entertainment. A long answer may signal care, hesitation or both. A short answer may reflect discipline or incompleteness. Refusal can demonstrate sound security judgment, while confident analysis can still end without a signature.
The broader lesson is that AI selection should resemble resilience planning. Buyers should examine performance during adverse conditions, including whether a model reads deeply, respects trust boundaries, escalates when blocked and completes valuable work. The Crucible League shows that frontier models can agree on what is happening yet produce materially different outcomes.
Firmulate also offers enterprises the same kind of wargame using a read-only export of their own business. Nothing writes back to real systems. That extends the central idea beyond a shareable quiz: before giving an AI workforce responsibility, test how it behaves when the week goes wrong.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Portable Solar Generator 300W Portable Power Station with 60W Solar Panel
- Portable Generator with 60W Solar Panel Included: with a big battery pack,...
- Multiple Charging outlets for camping gear with SOS Flashlight: with 2* 300W Max AC...
- Multiple Charging Optional, Solar Panel Charger 60W Included: ZeroKor portable power bank generator...
As an affiliate, we earn on qualifying purchases.
SOARAISE Solar Power Bank 48000mAh Wireless Portable Charger with 4 Cables
- Upgraded High-Efficiency 4 Solar Panels: Equipped with 4 premium solar...
- Massive 48000mAh Solar Power Bank: Featuring a high-capacity 48000mAh lithium-polymer...
- Built-in 4 Cable for Multi-Device Compatibility: Designed for multi-device charging, this...
As an affiliate, we earn on qualifying purchases.
