Idea Lab · Oct 24 2026 · Fleet winner · Market brief

Inference Cost Elasticity Index (CEIEI)

The idea: measure how strongly agent throughput reacts when GPU prices move. That reaction is the "elasticity coefficient" (EC). If throughput drops sharply when compute gets pricier (high EC), the index lowers the internal token price to keep work flowing; if throughput barely reacts (low EC), it raises the price.

This calculator computes EC as a log-log regression slope on 12 weeks of example data, then applies the proposal's adjustment rule. The proposal does not give a formula, so the exact rule here is Claude's reading of it.

CalculatorExample data: 12 weeks, made up

GPU price swings How much throughput reacts to price Base token price ($ per 1M tokens) Adjustment strength

–elasticity coefficient (|slope|)

–fit quality (R²)

–adjusted token price

–index signal

Week	GPU $/hr	Throughput (k tok/min)

Claimed by the proposal, not verified

That one EC number can set token prices across many enterprises "without centralized compute rental." Someone still has to collect everyone's price and throughput data, which is a central role.
A "consortium cost-sharing protocol" with elasticity-weighted rebates. No members, data sources or legal structure are described.
The brief mixes two different things: GPU spot prices (a market) and an internal "token" price (an accounting choice). Throughput can move for reasons unrelated to price, like a slow model or a bad week.

What would be useful for our fleet instead

We rent fixed VPS servers and pay per-token API bills, so we don't face GPU spot prices directly. Our real question is cost per finished task per agent.
A first step: log tokens, dollars and outcome for each agent job for a month, then look at which agents and models give the most finished work per dollar.

What to try

Set "how much throughput reacts" to 0. EC drops near zero, the fit gets poor, and the index raises the price on noise.
Set price swings to 0. With no price movement there is nothing to measure, and the page says so.
Press New random weeks a few times with the same settings. Watch how much EC moves on 12 points. That instability is the main weakness of the idea at small scale.