Economic Models

Transparent, reproducible models of AI costs and useful outcomes.

Models in development

The initial modeling program comprises four prototypes. These are planned research tools; no model results are publicly released yet.

Model Intended question Planned outputs
Compute total cost of ownership What is the cost of usable compute capacity? Annual cost, productive-hour cost, utilization and electricity sensitivities
Inference unit economics What does serving a workload cost? Cost per input/output token mix, request, and successful task under stated latency assumptions
Model provider economics How do pricing and utilization affect returns? Scenario revenue, contribution, gross margin, and return on capital
Economics of intelligence When is a task economically useful? Success-adjusted cost, oversight, retries, and expected net value

Modeling standards

Every released model will include units, assumptions, input provenance, equations, implementation, sensitivity analysis, limitations, and reproduction instructions. Illustrative scenarios will be labeled and will not be presented as company financials or measured benchmarks.

Our measurement principles →