AI agents consume five times more tokens than users on OpenRouter, analysis shows

Data collected by OpenRouter reveals that automated AI systems process nearly five times more information units (tokens) than direct human interactions on the platform. Since February, the volume of tokens attributed to these agents has grown by around 14 times, indicating a surge in the use of AI models to perform complex tasks.

How does high token consumption work?

When a user requests a task, an AI agent may consult the model multiple times: to plan execution, use tools, test results, and correct errors. Each step adds new data to the account, even if the user doesn’t send a new message. The agent re-presents instructions, incorporates new results, and requests new decisions from the model until the final answer is delivered.

Tokens represent the basic processing units of AI models. In text, a token can correspond to a word or parts of it. Both the query sent to the model and the generated response are counted, so a larger volume of tokens reflects a larger volume of content processed.

Anthropic data confirms global trend

A separate survey by Anthropic, released in June 2025, supports OpenRouter’s findings. According to the company, conversations conducted by AI agents consume approximately four times more tokens than human interactions in chatbots. Systems that use multiple agents can consume up to 15 times more tokens than a human user.

More processing doesn’t equal more value

An increase in token usage doesn’t necessarily mean greater productivity or financial return. The cost of each token varies depending on the model and the type of processing performed. Additionally, cached information can be reused in subsequent queries, reducing the amount charged.

OpenRouter’s activity chart measures only the volume of processing, without indicating how many tasks were completed successfully, how much time was saved, or what economic benefit was generated. Anthropic argues that multi-agent systems are only viable when the value of the task justifies the additional processing, and that the decisive economic metric is the cost of the completed task, not just the number of tokens consumed.

Frequently Asked Questions

• What does higher token consumption by AI agents mean?

  It means that queries made by agents process a larger volume of information than direct human interactions on the platform, without reflecting the number of agents or users present.

• What are artificial intelligence tokens?

  They are units used by models to process information. In text, they can correspond to words or fragments of words. Both the content sent to the model and the generated responses are included in the count.

• Why do AI agents use more tokens?

  Because a single request may require multiple queries to the model to plan, execute, test, and correct tasks, steps that occur without new messages from the user.

• Do more tokens mean more productivity?

  Not necessarily. The volume measures processing, but doesn’t indicate whether the task was completed correctly or whether the result justified the cost. Reused cached content may also incur a lower charge.