usage-based pricing (UBP)

basicwidespread

A payment model where you do not pay a fixed monthly fee, but only for what you actually used from the service, for example, based on the amount of text processed by AI.

In plain terms

Think of your electricity bill: you do not pay because the power is connected, but based on what the meter shows. Usage-based pricing applies this same principle to software and AI services: the more you use, the more you pay; the less you use, the less you pay.

In the AI world, the 'meter' usually counts tokens, meaning how much text the model has read and generated. Other providers bill based on the number of images created, minutes processed, or tasks completed.

This is different from the usual subscription, where the monthly fee is the same whether you use the service daily or not at all. Usage-based pricing seems fairer, but in exchange it is harder to predict how much the bill at the end of the month will be.

With the spread of AI agents, the question has become sharper: an agent works independently over many steps and can consume vast numbers of tokens during a single task, even though you issued only one request.

How it works

The provider specifies the unit of measurement for which you pay: this can be a token, image, audio minute, query, or completed task.
It assigns a unit price, for example per one million tokens. The more knowledgeable model is usually more expensive, and the response written by the model generally costs more than the text read.
The system automatically measures and logs every use, and consumption can usually be tracked by yourself on a control panel.
At the end of the period, the measured quantity multiplied by the unit price is invoiced, or deducted from a pre-loaded balance.
Most providers let you set a spending cap or alert so the bill does not run away.

An everyday example

A small business connects an AI-based customer service chatbot to its webshop and pays the provider based on the volume of processed text. In a quiet month, the bill is a few thousand forints, but during the Christmas rush it can be many times higher, as the fee follows the traffic.

Who is affected

entrepreneur, executiveAI cost is thus not a fixed item but moves with traffic, requiring different planning and monitoring.

developerModels accessed through an API are typically paid for by the token, so a wastefully written application is directly more expensive. → Code

employeeIf your company pays for AI tools based on usage, workplace use may have limits and can be audited. → Work & HR

everyoneMore everyday applications are introducing credits or monthly limits, after which you must pay or wait once they are exhausted.

A common misconception

MythUsage-based pricing is always cheaper than subscription because you pay only for what you use.

In realityFor small and occasional use it is indeed more favorable, but with intensive use a fixed monthly fee is often cheaper and more predictable. Before deciding, estimate your own consumption.

Limits and risks

The bill is hard to predict in advance because it is rarely visible from the outside how many tokens a request will cost.
An independently working agent or a faulty, repeating program segment can generate high costs in a short time, so always set a spending cap.
Units of measurement vary by provider (token, credit, task), making it difficult to compare offers directly.
The measurement is performed by the provider, and you can only verify it to a limited extent; for Hungarian text, the same content typically consists of more tokens than in English.

Similar, but not the same

subscription pricing – There you pay a fixed fee for a period regardless of actual usage; here the fee follows consumption.
outcome-based pricing – There you pay only for successfully achieved results (for example, a resolved customer complaint), while here you pay for the resources used even if the result is poor.
freemium modell – There, the basic service is free and you pay for extended features, not for the amount of consumption.

Related concepts

Broader

pricing model, business model

Narrower

token-based pricing, kreditrendszer

On the edge

token, subscription pricing, outcome-based pricing, API, AI-agent, cloud service, spending cap

A brief history

The principle originates from utility metered billing. In the software industry, cloud services made it common after Amazon launched its consumption-based storage and computing services in 2006. Large language model providers adopted this when they began billing their programming interfaces by tokens.

Where things stand

At the end of 2026, this is the default pricing for AI models accessed through programming interfaces (APIs). For consumer chatbots, the monthly subscription remains dominant, but hybrid models combining a monthly fee with a usage allowance and overage charges are becoming increasingly common.

Status: 2026 November

In the news

Appears in 3 Agent Age stories; the curve shows weekly occurrence in the news magazine.

Agent pull requests overtook human ones on GitHub in August as Copilot moves to usage-based pricingOct 5, 2026 · Code
Consumption-based martech brings surprise AI bills to CMO budgetsSep 5, 2026 · Marketing
AI coding tools move to usage-based pricing: Copilot credits and token ratesAug 31, 2026 · Code

Also known as

usage-based pricing · UBP · pay-as-you-go · consumption-based pricing · metered billing · usage-based pricing · usage-based pricing

Sources

OpenAI API Pricing

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.