Gartner Warns AI Agent Costs Will Rise 5x by 2028 — Even as Tokens Get Cheaper

The big picture

Most business owners following AI know that token prices — the basic unit of data models consume and generate — keep falling. The natural assumption is that your company's AI bill will shrink too. A new report from research firm Gartner, covered by VentureBeat, warns that assumption is wrong — and often the opposite is true.

Gartner calls this the "Inference Paradox": routing a task to an autonomous AI agent raises the real cost of running it by at least 5x compared to a basic chatbot interaction — and often much more as task complexity grows.

Why does this happen if prices are falling?

According to Will Sommer, a senior director analyst at Gartner, there are two combined drivers:

1. AI labs aren't banking the savings. As a model becomes cheaper per token, developers (OpenAI, Anthropic, Google, and others) don't pass that savings straight through — they reinvest it into more powerful, more expensive next-generation models. Gartner points out that Anthropic's Fable model costs more per token than its predecessor, Opus.
2. Agents consume far more tokens than chat. Gartner says completing a task via an autonomous agent requires 5 to 30 times more tokens than a basic chatbot question, since the agent must plan, verify, self-correct, and execute multiple sequential steps.

The result: the moment a business moves from "I asked the chatbot a question" to "I deployed an agent to complete an entire task," the real cost can be surprising.

Five practical steps for businesses

Gartner recommends five architectural changes for companies adopting AI agents:

• Invest in observability. You can't control a cost you can't see in real time.
• Tie spend directly to outcomes. Require every AI deployment to show measurable value against its cost (e.g., support tickets resolved — not just tokens consumed).
• Institutionalize smart routing. Default routine tasks to cheaper, efficient models, and reserve the "expensive and smart" models for tasks that truly need them.
• Set minimum standards upfront. Don't launch a new AI capability without stress-testing it against price swings and failure scenarios first.
• Leverage proprietary data. Domain-specific models fine-tuned on your own business data tend to be more reliable and efficient than generic frontier models.

A real-world example: the race for token efficiency

The same week, AI startup Reflection AI unveiled a new open-weight model called Beam, explicitly built to complete tasks using fewer tokens — not just a lower price per token. Per VentureBeat's report, on certain benchmarks Beam matched competing models' scores while using roughly a third of the tokens — a concrete example of exactly the kind of efficiency solution Gartner is recommending.

Why this matters for your small or medium business

If you're considering deploying autonomous AI agents — for customer service, data analysis, or operations — don't budget based on today's token price. Budget based on the expected cost of the full task, and track actual usage from day one. The difference between a business that profits from AI and one that gets blindsided by the bill usually comes down to monitoring and governance, not technology.