# AI Agent Inference Costs Could Rise More Than Fivefold Through 2028

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/33851
Published: 2026-10-17T08:29:29.000Z
Updated: 2026-10-17T08:29:29.000Z
Section: Technology

> A separate projection puts inference costs for a one-trillion-parameter model more than 90% lower by 2030 than in 2025.

Inference costs for an AI agent workflow are projected to climb more than fivefold through 2028 as tasks consume more tokens and use more capable models, even while the cost of processing tokens improves.

The projection measures the cost of completing a workflow, not the price of processing a single token. A workflow with more steps or a more powerful model can require a larger total inference bill, even as per-token costs decline.

The projected increase reflects three developments: foundational models are becoming cheaper to run, more powerful and expensive models are becoming viable for sophisticated applications, and those applications use more tokens than simple chatbot exchanges.

A separate comparison puts the provider cost of routing a task to an agentic reasoning model at least five times above a basic chatbot interaction. Costs can rise further as tasks become more complex. This model-choice comparison is distinct from the projection through 2028.

A March 25, 2026, projection estimated that inference on a one-trillion-parameter large language model could cost providers more than 90% less by 2030 than in 2025. That estimate covers a model of a specific size, while the other projection concerns the cost of completing an agentic workflow. They measure different things: the total bill for a task can rise even as the cost of running a particular model falls.

Agentic models can use five to 30 times more tokens per task than a standard generative AI chatbot. Tokens are units of data processed by AI models; the five-to-30-times comparison measures token use, not the cost of a task.

“Product leaders cannot rely on more efficient token economics to rationalize AI costs,” said Will Sommer, a senior director analyst at Gartner.

Inference tiering, routing and orchestration can help match complex work with suitable, more cost-efficient models. The more-than-fivefold projection is expressed as a multiple, without a dollar estimate or a stated baseline year.