Glossary · Everyday AI use
Usage-based pricing
Usage-based pricing is a billing approach in which the amount you pay depends on measured consumption of a service. For AI services, that consumption might include text processed, audio minutes, generated images, tool calls, or computing time, depending on what the provider chooses to measure.
Usage-based pricing: the explanation
The provider defines a unit of consumption and assigns a price to it. For a language model, it may count tokens, which are units of text rather than necessarily whole words. Input text and generated text can have different prices. Other services measure audio minutes or completed operations. Your bill therefore depends on how much you use, which resources are involved, and which features the task activates. A simple comparison requires matching the units and activities being charged.
This approach makes payment vary with activity. It can suit irregular workloads because you may avoid paying for a fixed amount of unused capacity. However, usage charges can also be combined with a subscription, minimum commitment, or prepaid credits. To estimate spending, test tasks that resemble what you will actually do. A short request on your screen might trigger several searches, intermediate responses, and additional calls, each consuming resources that may be included in the service's billing arrangements.
A conversation is not the same thing as a billing unit. An agent may retry a step, send a long conversation history, or consult several tools before finishing. Depending on the service, those additional steps can incur charges. Usage dashboards help explain the difference, but a budget alert does not necessarily stop spending. Check which limits are enforced, how errors are handled, and whether the displayed estimate can differ from the amount eventually charged for completed or attempted work.
A concrete example
You use an agent to summarize meeting notes stored in a folder. During a trial, you measure the resources used for a representative document. You find that longer files require more processing and that the agent sometimes retries an incomplete reading. Before processing the rest, you set a document size limit and check which budget controls actually stop work rather than simply notifying you about spending.
Why it matters
Usage-based pricing makes the cost of a successful task more useful than the advertised price of one unit. A cheaper service may require more steps or more corrections. Check the billing units, chargeable features, and available stopping controls. With an autonomous agent, monitoring retries and tool calls helps prevent a seemingly simple assignment from generating a bill that is difficult to anticipate.
Common misconceptions
- One request does not necessarily equal one billable unit, because an agent may make several model calls, consult external tools, or repeat unsuccessful steps before producing the final result that you see in the application interface.
- A budget alert does not guarantee that service stops at the stated amount, because some controls only send notifications while others enforce a spending limit under specific conditions that you need to check in the service settings.
Related terms: everyday ai use
- prompt
- A prompt is input supplied to a model to guide what it should produce or do.
- RAG (retrieval-augmented generation)
- RAG, or retrieval-augmented generation, combines information retrieval with a model’s ability to generate an answer.
- vector database
- A vector database stores and searches numerical representations called vectors.
- semantic search
- Semantic search retrieves material according to its likely meaning, beyond matching the same words.
- coding assistant
- A coding assistant is an AI tool that helps you write, understand, change or test computer programs.
- no-code
- No-code is an approach to building applications or automations through visual interfaces and settings, without directly writing code for the supported functions.
- text-to-speech
- Text-to-speech is a technology that converts written text into machine-generated speech.
- speech recognition
- Speech recognition is a technology that turns spoken audio into text a computer can process.
Where this site covers it
The term appears on 10 pages of this site. A few of them:
Same family
Go further
Frequently asked questions
What is a billable token?
A token is a unit used to represent text processed by a model. It might be a word, part of a word, or punctuation. The count depends on the text and the tokenizer used by the model.
Can I predict the cost?
You can estimate it by running representative tasks and measuring their full consumption. Include long documents, retries, and connected tools. The estimate still depends on the agent's actual behavior and the pricing terms that apply to your service.
Are unsuccessful requests charged?
That depends on the service and the kind of failure. An unusable answer may still have consumed resources even though your task was not completed. Check the terms and usage details rather than assuming every unsuccessful attempt is free.
Neutrality. No affiliate links. No vendor pays to be reviewed, rated or ranked. Reviews are directed by François Chollet, neutral and factual, and every data point is dated and sourced. How we score