Cost per Task

Cost per Task is a metric from economics. It answers a simple question: What does it cost to have a single completed task carried out by a computer program? A task might be, for example: answering an email, describing a photo, checking an invoice for errors. You add up everything that arises for this one task and divide it by the number of tasks completed. The result is an amount, often just fractions of a cent. The English term for this is Cost per Task, and it frequently appears in this form in business reports.

Why companies are suddenly calculating in tasks

In the past, software was compared based on the subscription price per user and month. For programs that carry out tasks themselves, this no longer fits. A user might submit ten tasks a month or ten thousand. The effort for the provider thus differs by a factor of a thousand, but the price does not. That is why many companies now bill per task completed.

The second reason is the comparison with human labor. If a case worker needs twenty minutes to review an application, that can be converted into money. If the same review by a program costs two cents, the decision is obvious for a company. Cost per Task is thus the number that determines whether automation is worthwhile or not.

At the same time, the metric is a warning signal for providers. Some companies charge a fixed monthly price but have fluctuating costs per task. If individual customers use the system very intensively, the provider loses money on them. This exact problem forced several providers of AI coding assistants to raise their prices in 2024 and 2025.

What goes into the calculation

The largest item is usually the pure computing time. A language model — that is, a program that continues texts and answers questions — processes the text in small units. These units are called tokens and correspond roughly to a syllable. Providers charge a price per thousand or per million tokens. Anyone who knows how many tokens a task consumes can calculate the costs fairly precisely.

The catch: a task rarely consists of a single request. Modern systems call the model multiple times in succession, check their own result, look up data, and correct themselves. One request quickly turns into fifteen. On top of that come database access, storage space, and the costs for tasks that go wrong and have to be repeated. Anyone who only looks at the token price significantly underestimates the real costs.

A common mistake is to confuse Cost per Task with the price per token. The two figures can develop in opposite directions. Token prices have fallen sharply since 2023. Nevertheless, for some applications the cost per task rose, because the systems compute far more intermediate steps for better results. Cheaper per unit does not automatically mean cheaper per result.

Where the figure appears in news and products

It is most visible in price lists. Providers of customer service software often charge a fixed amount per resolved inquiry, for example one dollar per case. This is Cost per Task as a sales argument: the customer only pays for results, not for attempts. This billing approach is also becoming established in translation services and automatic image editing.

In business news, you encounter this metric when it comes to the profit margins of AI companies. Analysts then ask whether a provider earns more per task than the task costs it. For many young companies, the answer is still no. They finance the difference with investor money and hope that computing power will continue to get cheaper.

For you as a user, this has an indirect effect. If a free service suddenly shortens answers or introduces a daily limit, this calculation is usually behind it. The provider is simply lowering its cost per task. Anyone who knows the metric understands such changes as a business decision rather than a technical glitch.