Gartner: AI agent costs rise 5x by 2028. Tier your models

Token prices keep falling, and your AI agent bill can still go up. Gartner forecasts that AI agent costs per workflow will rise more than fivefold through 2028. This week VentureBeat added survey data showing how few companies have controls in place. If you run agents in your business, the per-token price is the wrong number to watch.

What actually happened

Gartner calls it the "Inference Paradox." Sending a task to an agentic reasoning model costs the provider at least 5x more than a basic chatbot exchange, and often more as the task gets harder. In VentureBeat's October 16 report, Gartner senior director analyst Will Sommer gives three causes:

• Labs spend their efficiency gains on bigger models. The savings do not reach your invoice as lower prices. They ship as pricier frontier tiers.
• Agents use 5 to 30 times more tokens per task than a chatbot doing the same work.
• Competitive pressure moves companies onto harder use cases, which need bigger models and more tokens.

"Token costs aren't falling; they're widening," Sommer told VentureBeat. A September VB Pulse survey of 109 respondents found that 21% of companies using an orchestration platform have nothing that automatically stops agent token spend. Only 31% route routine, high-volume agent work to cheaper models.

Why it matters for your business

Big companies can absorb a 5x surprise for a quarter. A ten-person business finds it on the credit card statement. The fix is not to avoid agents. It is to stop running every agent step on the most expensive model.

Gartner's main recommendation is inference tiering: send each query to the cheapest model that can do it, and don't let the frontier model be the default for routine work. In practice, a cheap model classifies, extracts and drafts. The expensive model only gets the steps where reasoning changes the result. Your evals decide where that line sits, not a guess.

Then add the controls the 21% are missing: a hard spend cap per agent, cost per task logged next to outcome per task, and a review of any workflow whose cost rises faster than its value. Sommer's point stands: "Enterprises cannot optimize what they cannot observe."

Sources: Gartner, VentureBeat.