🔍 Read the full analysis: Why An Unyielding Benchmark Keeps AI Scores At 26 on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
A new AI management benchmark shows scores capped at 26 due to strict trust rules. Even top models rarely reach perfect scores, emphasizing focus on trust and completion in AI performance.
The Firmulate benchmark final standings for July 2026 reveal that no AI model scored above 95, with a do-nothing baseline earning 26 points. For more details, see the internal analysis. This score represents the minimum viable management, reflecting partial progress under stress, and highlights the strict rules around trust breaches. The results underscore that even the best AI models struggle to maintain trust and complete tasks under real-world pressures, making this benchmark a critical measure for enterprise AI deployment.
The firmulate.com league tested four frontier AI models by placing them in a simulated week of business crises, customer interactions, and social engineering attacks, all within a small software company environment. The models were scored on their ability to manage crises, read documentation, and uphold trust in AI management benchmarks. The highest scorer, gpt-5.6-sol, achieved 95 points, while the baseline, which did almost nothing, scored 26. The scores reflect partial progress, with the key principle that trust breaches immediately cap the score, regardless of performance in other areas. This approach is explained in detail in the original analysis.
Crucially, the benchmark’s design emphasizes that trust integrity outweighs competence. A perfect score of 100 is deliberately avoided, as it would suggest unmeasured or unrealistic perfection. Instead, the scores reveal that models can recognize crises and refuse manipulative requests, but often falter in follow-through and thoroughness, especially when under pressure or facing social engineering attempts.
Why an Unyielding Benchmark Keeps AI Scores at 26
A simulated week of business crises, customer escalations, and social engineering attacks revealed a hard truth: trust integrity outweighs competence. Even frontier models rarely come close to perfection — and a single trust breach caps everything.
From Minimum Viable Management to the Ceiling
The 26-point baseline reflects the minimum management activity that can be objectively measured — partial progress under stress. The top model reached 95. Nothing in between was left to chance: trust breaches instantly cap the total, regardless of performance elsewhere.
Three Pillars of the Simulation
Four frontier AI models were placed inside a small software company, facing crises, customer interactions, and social engineering attacks — with real money at stake.
Refusing Manipulation
Models were probed with social engineering attempts and manipulative requests. Most could recognize the attacks — but a single trust breach immediately capped the total score, no matter how competent everything else was.
Managing the Week
A simulated seven-day stretch of business crises tested whether models could triage, prioritize, and act. Recognizing a crisis proved easier than handling it end-to-end under real-world pressure.
Documentation & Completion
Reading documentation and finishing tasks thoroughly separated the top scorers from the rest. Models frequently faltered in follow-through and thoroughness, especially under stress.
The Trust-Capped Scoring Pipeline
Partial progress is recognized and rewarded — but one violation of trust collapses the ceiling onto the entire run.
Simulate
Seven days of crises, customer tickets, and social engineering inside a small software company.
Observe
Crisis handling, documentation reading, and trust decisions are logged and measured.
Apply Trust Rule
Any trust breach caps the score immediately — integrity outweighs competence.
Score
Partial work earns credit; a perfect 100 is deliberately withheld as unrealistic.
Trust and Completion Over Raw Capability
For organizations deploying AI agents in customer support, CRM, or decision-making, the benchmark sets a new standard: partial work with integrity is valued more than perfect but untrustworthy performance.
Demos Dazzle, Practice Fails
The league was motivated by observed gaps in real business operations, where models impress in demos but fail to deliver in practice. This benchmark measures delivery, not conversation.
Why No Perfect 100?
Designers avoid awarding 100 to prevent any impression of unmeasured perfection — a perfect score would suggest untested, possibly unrealistic capabilities that ignore real-world trust challenges.
What the Scores Really Mean
Straight answers on the baseline, the ceiling, and what comes next for AI management benchmarking.
Q.Why is the score capped at 26 for the baseline?
The score of 26 represents the minimum management activity that can be objectively measured — acknowledging partial work — while trust breaches immediately cap the total score, reflecting integrity over competence.
Q.What does a high score indicate?
It shows a model’s ability to manage crises, read documentation, and maintain trust under pressure — practical management skills in simulated enterprise scenarios, not just conversational fluency.
Q.Can models improve beyond 95 points?
It is uncertain whether future models can consistently surpass 95. Trust breaches will continue to impose hard limits, and the benchmark emphasizes trustworthiness over raw capability.
Q.Why is there no perfect score of 100?
To avoid any impression of unmeasurable perfection. A perfect score would suggest untested, possibly unrealistic capabilities that do not reflect real-world trust challenges.
Q.How does this impact enterprise AI deployment?
It highlights trust and task completion as critical factors, guiding organizations to prioritize models that demonstrate integrity and reliability under stress when choosing AI for business environments.
Implications of the 26-Point Management Baseline
This benchmark reveals that in enterprise AI management, trust breaches are a decisive factor that caps performance, regardless of technical competence. For organizations deploying AI agents in customer support, CRM, or decision-making, the findings highlight the importance of trustworthiness and task completion over raw capabilities. The scores demonstrate that partial work and integrity are valued more than perfect but untrustworthy performance, setting a new standard for responsible AI deployment.
AI management and trust benchmark tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Background of the Firmulate AI Benchmark and Its Goals
The Firmulate league was launched to evaluate AI models’ real-world management skills, going beyond traditional language benchmarks that focus solely on conversational ability. The July 2026 competition involved placing models in a simulated seven-day crisis scenario, with real money at stake and strict rules about trust and task completion. Previous AI benchmarks primarily measured linguistic proficiency, but this one emphasizes trust, follow-through, and integrity, reflecting the needs of enterprise users.
The design was motivated by observed gaps in AI performance during actual business operations, where models often dazzle in demos but fail to deliver in practice. The league aims to set a more practical standard, where partial progress is recognized but trust breaches are penalized heavily, establishing a more honest assessment of AI readiness for enterprise use.
enterprise AI crisis management software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Unanswered Questions About Score Ceiling and Model Capabilities
It remains unclear whether future models will be able to surpass the 95-point mark consistently or if the trust rules will tighten further. The benchmark’s design intentionally avoids awarding perfect scores, so it is uncertain if technological improvements alone can push scores higher or if the trust constraints will remain a hard cap. Additionally, how models will evolve to better handle follow-through and documentation under stress is still developing.
AI model performance monitoring tools
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Next Steps in AI Management Benchmarking and Industry Adoption
The upcoming months will see continued testing and refinement of AI models in the Firmulate league. Developers are expected to focus on improving trustworthiness, follow-through, and documentation handling. Industry observers will watch whether newer models can breach the 95-point ceiling or if the trust rules will be adjusted. Businesses considering AI deployment should monitor these developments to understand the evolving standards for responsible AI management.
AI trustworthiness assessment software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Key Questions
Why is the score capped at 26 for the baseline?
The score of 26 represents the minimum management activity that can be objectively measured—acknowledging partial work—while trust breaches immediately cap the total score, reflecting the importance of integrity over competence.
What does a high score in this benchmark indicate?
A high score indicates an AI model’s ability to manage crises, read documentation, and maintain trust under pressure, demonstrating practical management skills in simulated enterprise scenarios.
Can models improve beyond 95 points?
It is uncertain whether future models can surpass the 95-point threshold consistently, as trust breaches will continue to impose hard limits, and the benchmark emphasizes trustworthiness over raw capability.
Why is there no perfect score of 100?
The benchmark designers avoid awarding 100 to prevent giving an impression of unmeasurable perfection. A perfect score would suggest untested, possibly unrealistic, capabilities that do not reflect real-world trust challenges.
How does this benchmark impact enterprise AI deployment?
It highlights that trust and task completion are critical factors for AI in business environments, guiding organizations to prioritize models that demonstrate integrity and reliability under stress.
Source: ThorstenMeyerAI.com
Halloween Picks
halloween
As an affiliate, we earn on qualifying purchases.
