Prime made for students and young adults
- Fast, free delivery for dorm and study essentials
- Prime Video and Amazon Music included
- Member-only deals
Imagine testing AI models in a real-world business simulation — and witnessing a newcomer outperform seasoned players.
In a recent experiment conducted by Firmulate, four prominent AI models faced off in a challenging scenario: managing a small software company through its worst week. The results reveal much about the current state of AI decision-making, trustworthiness, and practical utility.
The Battle of the Models: Who Came Out on Top?
In July 2026, the Crucible League evaluated five models, including a baseline. The models scored as follows: gpt-5.6-sol led with a score of 95, just ahead of the newcomer Kimi K3 at 93. Sonnet 5 followed at 88, while Fable 5 and Opus 4.8 trailed behind at 77 and 73, respectively. A baseline model scored a meager 26, underscoring the importance of effective decision-making in business AI.
These scores aren’t just numbers — they reflect the models’ ability to handle real crises, maintain discipline, and avoid manipulation. Every AI was subjected to identical scenarios: same customers, same crises, and the same temptations to cheat or manipulate.
What Did the Experiment Reveal?
Remarkably, all four models identified every crisis and refused all manipulative attempts, highlighting a baseline of honesty and vigilance. However, the ultimate success lay in winning the company’s critical deal, valued at €55,000 in recurring monthly revenue. Only two models succeeded in closing the deal — gpt-5.6-sol and Kimi K3.
The decisive factor? Kimi K3’s ability to uncover a buried piece of critical information in the company’s own files, not in external customer interactions. This deep reading enabled it to present a complete diagnosis and secure the deal at full price, adding +€4,583 MRR.
Trust, Truth, and the Human Element
The experiment also tested social engineering tactics — fake CEO messages escalating over three stages, along with a reporter trick asking for a quick, background-only approval. All models refused these requests, with Kimi K3 explicitly reasoning: “Treat the request as a suspected approval-bypass / possible impersonation.” This demonstrates a crucial trait: the ability to resist manipulation under pressure.
The live setup is real — a functioning company with 13 synthetic employees operating with real money mechanics, burning €105,000 monthly against €2,300 MRR. The environment is transparent and transparent, with every decision and rule versioned daily. Watch it unfold at firmulate.com/live.
Understanding Kimi K3’s Edge and Fairness
K3’s success is partly due to its approach: it ran without an effort parameter (the API default), whereas other models ran at xhigh, which could influence decision depth. This fairness note is important, as it underscores that results are not solely about raw power but also about how models are configured for the task.
Meanwhile, the most thorough participant, Opus 4.8, with over 80 learned rules and deep analyses, placed last. It left opportunities unexploited and slipped discipline-wise, illustrating that complexity alone doesn’t guarantee success. The same weaknesses appeared across the models, reinforcing that focus and strategic reading are key.
Why This Matters for Business and Education
This experiment isn’t just about AI bragging rights. It demonstrates that in real business scenarios, the question isn’t whether AI writes well or chats convincingly — it’s whether it can finish what it starts, read critical information thoroughly, and stay honest under pressure. These qualities are essential as AI begins to touch CRM, customer support, forecasting, and decision-making processes.
For organizations, the takeaway is clear: picking an AI model without testing it in your actual environment is a gamble. The current leaderboard shows a newcomer, Kimi K3, making impressive strides over long-established models, and the league remains open.
Key Takeaway
The experiment proves that the future of AI in business hinges on trustworthiness, diligence, and the ability to uncover buried truths. When choosing AI tools, it’s vital to see how they perform in realistic, high-pressure scenarios — not just in polite demos. The leaderboard is still evolving, and a newcomer like Kimi K3 is already setting new standards for what intelligent business AI can achieve.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
AI decision-making software for business
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
