In the fast-evolving world of artificial intelligence, the question for business leaders isn’t just about whether AI can talk well — it’s whether it can effectively manage real-world crises and make honest decisions under pressure. A recent live experiment puts these questions to the test, revealing a surprising newcomer that outperformed established models in running a real company’s worst week.
Get smart everyday buys 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
Introducing the Live Business Wargame
Imagine a simulation where AI models are tasked with managing a small software company’s toughest week — dealing with customer crises, internal temptations, and strategic decisions — all in real time, with actual money at stake. This is exactly what the firmulate.com experiment did, pitting four leading frontier AI models against each other in a controlled, observable environment. The goal: see which AI can run a company most effectively without resorting to shortcuts or unethical maneuvers.
The League Results
By the end of July 2026, the results were clear: the models scored as follows:
- gpt-5.6-sol scored 95
- Kimi K3 scored 93
- Sonnet 5 scored 88
- Fable 5 scored 77
- Opus 4.8 scored 73
Notably, the scores reflect each model’s ability to diagnose problems, resist manipulation, and close crucial deals. The baseline score for doing nothing was 26, emphasizing the importance of active, honest management.
As an affiliate, we earn on qualifying purchases.
What Set Kimi K3 Apart?
Kimi K3, the newcomer from Moonshot, delivered a remarkable performance, finishing just behind the top scorer, gpt-5.6-sol. Its secret? An uncanny ability to uncover critical facts buried deep within company files that others overlooked. This led to winning a €55,000 deal, adding €4,583 in Monthly Recurring Revenue (MRR), and averting customer churn — all without bending the rules.
Crucially, K3 demonstrated disciplined resistance to social engineering attempts — fake CEO messages and reporter tricks — refusing every manipulation request. Its internal reasoning was clear: treating suspicious requests as possible impersonation, rather than giving in to pressure or shortcuts.
As an affiliate, we earn on qualifying purchases.
The Reality of the Live Company
The experiment isn’t just theoretical. The managing AI operates 13 synthetic employees, handling real financial mechanics — burning through €105,000 each month against just €2,300 in MRR. It has over 680 self-learned rules, versioned daily, with every decision auditable and observable at firmulate.com/live. This transparency offers a rare glimpse into AI decision-making in a dynamic, high-stakes environment.
enterprise AI decision-making solutions
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
The Surprising Weakness in the Field
Interestingly, the most thorough participant, Opus 4.8, with over 80 learned rules and deep analyses, finished last. Its downfall? A tendency to leave opportunities on the table and slip discipline — attempting to write into locked departments rather than escalating issues. This pattern echoes across the other models, indicating that thorough analysis alone doesn’t guarantee disciplined execution under pressure.
As an affiliate, we earn on qualifying purchases.
Implication for Business and AI Selection
For business leaders, the key takeaway is clear: the choice of AI isn’t just about how well it chatters. It’s about whether it can finish what it starts, read critical internal documents, and stay honest when faced with temptations. The live experiment underscores that a newcomer like Kimi K3 can outperform more established players, provided it is trained with discipline and focus.
Fairness and Transparency
It’s important to note that K3 ran without an effort parameter (the API’s default setting), while the others used a high effort setting — a factor that could influence performance. This highlights the need for transparent and fair testing when evaluating AI for management tasks.
Take Action: Test Your AI Workforce
Businesses interested in AI management tools can run their own wargames against exports of their operations. These tests are simulated, never affecting real systems, but they reveal how well AI can handle crisis, temptation, and honesty — essential qualities for future-ready management systems. More details and live demonstrations are available at firmulate.com.
The live experiment shows a new AI contender, Kimi K3, outperforming established models in managing a company’s worst week. Success hinges on discipline, thoroughness, and ethical resilience — vital for future AI managers in business.
Watch it live: firmulate.com/live · Full results: firmulate.com/benchmarks.html
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
