The denomination

A Common Compute
Basis Token.

An economic reference for AI consumption—not a particular model, a particular GPU, or a universal measure of intelligence.

General mathematical model · v0.1 conceptExplore the research

01 / The common denominator

A Common Basis
for AI Economics.

CCBT asks a simple question: what is the average market price of a token in the AI consumption actually taking place?

The two inputs are aggregate market usage and each model token’s open-market price. The models the market uses most contribute the most weight. As that consumption shifts, the common reference changes with it.

The model, in plain English

The price of the tokens
the market actually consumes.

Give each model’s token price a weight equal to its share of aggregate market token usage. Add those weighted prices. The result is the price of one common token for that market and period.

Input 01 / The market’s consumption

Aggregate market usage.

How many qualifying tokens of each model the whole defined market consumes—not one buyer’s usage or preferred basket.

Model’s market tokensAll market tokensUsage share
Input 02 / The market’s prices

Open-market token prices.

What a token from each qualifying model costs in the market, expressed on a consistent pricing basis—not the cost of a particular GPU or a score for model quality.

Model tokenObservable standardized price
Put the two together

CCBT priceSum of (model price × market token usage)Total qualifying market tokens consumed

The same idea, in mathematics
PCCBT,t=Σi pi,t qi,tΣi qi,t=Σi wi,t pi,t
p = standardized price per model tokenq = model-token usage across the defined marketw = q ÷ total market-token usage

Market usage sets the weights. Your allocation comes later. A model serving 40% of qualifying market tokens carries 40% of the weight. The result is usage-weighted, not an equal average of available models.

Read the methodology
Once the common unit is defined

One price to observe.
One denominator to use.

The dollar price of CCBT describes the common reference. Expressing model costs in CCBT gives businesses a common cost basis for their own analysis and allocation.

$/CCBTPrice the common unit.Observe how the price of market consumption changes.
CCBT / model tokenPrice a model against it.Model-token dollar price ÷ CCBT dollar price.
A common cost basis. A business-specific value judgment.

The market prices the cost.
You measure the benefit.

CCBT supplies a common economic denominator. Whether a model’s premium earns its keep depends on the value it creates for your workload—the numerator in your cost-benefit analysis. The standard does not assign a universal score to intelligence or completed work.

Beyond a benchmark

From price discovery
to financial markets.

Cost analysis & business planningProcurement & model allocationFuture consumption commitmentsFinancing & price-risk transfer

The common denominator is a basis for the financialization of AI—not only a financing product. Financial applications depend on contracts, counterparties, and market development.

Definition

A CCBT is the proposed price of a hypothetical universal AI token whose composition reflects which qualifying models a defined market actually uses and what those model tokens cost.

Tokens are the common commercial unit by which model consumption is measured and priced, but they are not standardized economic units. Each model has its own tokenization, architecture, computational requirements, pricing structure, and performance characteristics.

CCBT does not reconstruct all underlying compute costs or measure intrinsic model quality. It takes observable model-token prices as its starting point and establishes a common denominator against which market participants can evaluate capability, efficiency, and usefulness for themselves.

Basic Formula

For a defined CCBT class c at time t:

PCCBT,c,t = Σi=1…n wi,tPi,t

Where Pi,t is the observable standardized token price of qualifying model i, wi,t is its share of qualifying utilization, and n is the number of models in the relevant basket.

wi,t = Ui,t ÷ Σj=1…n Uj,t    and    Σ wi,t = 1

Equivalently:

PCCBT,c,t = (Σ Ui,tPi,t) ÷ (Σ Ui,t)

In plain language: observe each qualifying model’s standardized token price, determine its share of relevant utilization, multiply price by utilization weight, and sum the results.

Usage means aggregate market consumption. The weight is based on qualifying tokens consumed across the defined market, not a customer’s preferred model allocation. Prices are observable model-token prices on a consistent economic basis; the implementation specifies how different price types are normalized.

Index Construction, Numeraire Purpose

The mathematics are conceptually comparable to a basic price index: heterogeneous prices are combined using weights that represent their relative importance in a defined market.

The intended purpose is different. CCBT is not merely proposed to describe whether inference has become more or less expensive. The basket constructs a standard denomination in which model costs themselves can be expressed. A model might trade at 0.8 CCBT, 1.0 CCBT, or 1.2 CCBT relative to its class.

CCBT does not decide whether a 20% premium is justified. The user compares that common cost difference against the value generated for its particular workload. The market determines price; the user determines value.

Assumption of Liquid Compute

The theoretical basis begins with a simplifying assumption: underlying compute is treated as fungible, liquid, and globally available at a prevailing market price. Real markets will not perfectly satisfy this assumption.

Pactual = PCCBT + Basis

Geography, latency, hardware scarcity, capacity guarantees, data residency, availability, provider quality, contractual terms, and credit could be expressed as premiums or discounts around the common reference.

CCBT Model Classes

There need not be one basket for all AI. Economically distinct markets can have separate specifications with common methodology but different eligibility rules.

CLASS 01

GenAI Inference

Broad qualifying general-purpose inference.

CLASS 02

Frontier

Qualifying frontier general-purpose models.

CLASS 03

Open Inference

Qualifying open-model inference.

CLASS 04

Coding

Software-development workloads.

CLASS 05

World Models

Qualifying world-model workloads.

Dynamic Composition

CCBT is intended to represent the market that exists, not a fixed collection of models. Weights change as utilization changes. New models can enter after satisfying eligibility and liquidity rules; obsolete or immaterial models can leave.

This is essential to long-term measurement. The denomination does not require today’s models to remain relevant or today’s hardware to remain in service.

Future AI Consumption

A contract tied to a contemporary GPU or model can become obsolete with its underlying technology. CCBT is proposed to separate the economic quantity of future AI consumption from the particular model through which it ultimately occurs.

A deliverable structure could allow a buyer to contract for a CCBT purchasing balance and allocate it among qualifying models at delivery according to prevailing conversion prices. This is a separate contract-design claim from the benchmark formula itself: CCBT is the denomination; each trade still needs an explicit settlement and delivery specification.

A quantity contracted for future delivery references the market then in effect. A 100M CCBT balance is not a frozen claim on today’s models: at delivery, the qualifying universe and contemporary conversion prices determine how it can be allocated. The market’s weights construct the reference; the buyer chooses its own consumption mix.

Open Methodology Questions

  • PricesHow input, output, cached, batch, committed-volume, and bundled prices should be normalized.
  • UtilizationWhich data sources credibly measure qualifying usage and how sampling bias should be addressed.
  • EligibilityHow models enter, leave, or move between economically distinct classes.
  • TimingMeasurement windows, publication frequency, lags, rebalancing, and restatement policy.
  • GovernanceOversight, conflicts, manipulation resistance, auditability, licensing, and exceptional-market procedures.
  • ContractsCash versus physical settlement, delivery rights, provider obligations, basis adjustments, credit, and regulation.

CCBT does not try to solve every variable that makes AI economics complicated. It proposes a common denomination through which the market can price their combined result.

From Standard to Research Implementation

The general model defines the economic object: a usage-weighted price for qualifying market consumption. A specific implementation defines the universe, observable price convention, aggregate usage sources, measurement windows, and publication rules.

The theoretical target is the price of what the whole defined market consumes. Observations from one provider or broker must be described in terms of their actual coverage; they are not automatically the entire market.

Dated observations, source records, implementation status, and release notes now have their own home in Research. Fictional demonstrations remain separate.

Research implementation See the core model
Contribute to the standard

Work on the methodology.

Discuss pricing conventions, representative usage data, eligibility, or publication rules.

Discuss the research