AIME | AI Model Economics
Discover how model selection, consumption, performance and risk affect enterprise AI economics. AIME is a specialized capability of Consumption Model Economics inside MOM-001 Capital Discipline. It is not a separate Micro Operating Model.
Are we using the right intelligence, for the right work, at the right cost and acceptable risk?
The economic opportunity, in thirty seconds
AI model consumption across eight workloads, scored on cost per successful outcome: tokens for every attempt, plus the cost when a person has to finish the work. Data as of Monday, 7:00 a.m. (fictional).
As intelligence becomes cheaper to consume, is your enterprise becoming more economical to operate?
Spending by provider and model
Risk and governance status
Annualized at the current monthly run-rate. Recommendations are proposals; nothing changes without approval. Model utilization by workload is shown as share of AI spend.
The cheapest model is not necessarily the most economical model
Choose a workload. Every model is priced from its provider's published list price, then scored on what one successful outcome costs. Change the ranking to see how far the two orders differ.
Change the work, the requirements, or the rules
Select a workload and change its consumption or operating requirements. Changes carry through the dashboard, the simulator and the recommendations.
Consumption assumptions
Operating requirements
Approved provider restrictions Enterprise-wide
Castle Pines Mutual's vendor-risk approvals (fictional). A provider that is not approved is excluded from every workload.
Monthly cost under alternative models
Twelve months forward, with growth
Premium for every task, or an economically optimized portfolio?
Scenario 1 puts every workload on one premium model. Scenario 2 assigns each workload the approved model with the lowest cost per successful outcome that meets its quality, latency, data and reliability requirements. Today's portfolio is shown between them.
From saving to net economic benefit
Estimated is not validated
Never assume two models produce equivalent results because they process the same tokens. Acceptance rates here are illustrative category assumptions until a shadow evaluation on the enterprise's own work replaces them.
BUY–HOLD–SELL recommendations
AIME's analysis becomes Capital Discipline verdicts, each with its economic rationale, evidence, risk, contract implications, financial impact, required approver, and rollback path. Approving here records a decision in this demonstration only.
The market moves whether or not your contracts do
AIME watches the external model market as well as the enterprise's own consumption: open-weight and closed-weight trends, price changes, new models and retirements, capability changes, hosting economics, licensing and availability. That intelligence feeds the existing Mark-to-Market and Scenario Economics lenses of Consumption Model Economics. It is not a seventh lens.
Open-weight versus closed-weight token consumption
Model pricing across the market today
Blended list price per 1M tokens, weighted 10 input to 1 output, from verified provider pages. A price is where the evaluation starts, not where it ends.
Market events and the lens they feed
Is Castle Pines becoming more economical to operate?
A market change triggers a new evaluation Fictional event
Castle Pines processes underwriting documents on a premium proprietary model. A new model becomes available at a substantially lower price. Inject the event and watch AIME evaluate it before anything changes.
Each feed shows how it is sourced today and what replaces it in production. The demonstration does not depend on any live external feed.
How AIME recommendations could inform routing
AIME is the economic decision engine. Routing is the enterprise's own infrastructure, its AI gateway, and it routes only what accountable people have approved. This page operates no production routing and calls no model.
Routing policy in force
Simulated request stream
Sixteen requests drawn from a fixed seed, so every run is reproducible. Approve BUY lines above and run again: only approved changes are routed.
An open licence is not a zero operating cost
Open-weight models can be downloaded and run privately. The licence may cost nothing; the GPUs, power, people, security and spare capacity do not. Because a cluster costs the same busy or idle, AIME allocates its full cost to the workloads that use it.
Infrastructure and operations
Monthly cost of the private cluster
Cost per million output-equivalent tokens, by utilization
Where self-hosting breaks even
Every number is traceable or labeled
Provider prices were checked on 25 October 2026 against each provider's own pricing page, except where marked unverified. Prices change; a production deployment would read them from provider agreements and usage APIs. Acceptance rates are illustrative assumptions by model category, not measured benchmarks of any named model. All enterprise data is fictional.
Provider list prices, USD per 1M tokens
The only place providers and models are named. Everywhere else, the analysis uses anonymized labels by category (Provider A–F, Premium, General-purpose, Low-cost). It does not recommend any named vendor.
First-pass acceptance by model category Illustrative · editable
Share of attempts that pass the workload's acceptance test without rework. Replace these with results from your own evaluations.
Method
Cost per attempt = (uncached input × input price + cached input × cache price + output × output price) × environment uplift.
Expected attempts = (1 − (1 − a)k) ÷ a, where a is first-pass acceptance and k the attempts allowed before a person finishes the task.
Cost per successful outcome = cost per attempt × expected attempts + (1 − a)k × cost of human completion.
Self-hosted: monthly cluster cost (compute or depreciation, power and hosting, staff, security tooling) allocated by output-equivalent tokens; input tokens weighted at 0.25 and cached input at 0.05 of an output token. Workloads share the cluster only where the shared cost beats their best approved API option.
A transparent demonstration cost model. BlueHour's production scoring, weighting and orchestration are proprietary and not shown.
Policy and costs
Data-handling policy (fictional): Public and Internal data may use any approved provider API; Confidential data requires a data-residency endpoint with a published price, or self-hosting; Restricted data requires self-hosting.
The entry point is AI model spending. The opportunity is enterprise-wide Capital Discipline.
An enterprise often begins with one narrow question: are we overspending on AI model consumption? Answering it properly raises the next ones. Are we using the right infrastructure? Are we overcommitted to GPU capacity? Are our cloud consumption arrangements economically justified? Are our SaaS and AI contracts still competitive? Are we allocating capital to the technologies that create the greatest enterprise value? Are our AI operating decisions consistent with our risk tolerance?
AI Model Economics → Consumption Model Economics → Capital Discipline → Operating Model Modernization
AIME's recommendations can inform other Micro Operating Models through governed interfaces: MOM-305 Operating Risk, where model and provider changes need containment and recovery paths; MOM-505 Governance & Accountability, where approval thresholds and decision rights are set; and Operating Model Stress Testing, where AIME's price, provider and model-retirement scenarios become stress inputs.
More intelligence requires more architecture. More consumption requires more Capital Discipline.