Open models can do 90% of agentic work at a twentieth of the cost of a frontier LLM: Gartner

Gartner, a leading analyst firm, predicts that open-weight models will be responsible for executing 90% of routine steps in enterprise agentic AI workflows, at a fraction of the cost of frontier LLMs. According to Gartner's Mark Petty and Raymond Paquet, open models can handle 90% of the routine tasks involved in an agent workflow - classification, tool calling, passing, validating, and reconciling data - at parity with closed models, but at a quarter of the price.

They estimate that open-weight models can perform hundreds of these routine steps for every one that requires frontier reasoning. While current open models may be more verbose and retry more often, leading to a saving of five to six times per task compared to a frontier model, the analysts consider this a welcome benefit, rather than the primary goal.

The cost savings stem from moving routine steps to open models, which helps maintain control over enterprise data and operations. By 2030, Gartner anticipates that 90% of enterprise operational agent workflow steps will be executed on open-weight models. They recommend that enterprises adopt domain-specific open-weight models, as they perform well in scenarios where a small number of highly stable steps run millions of times per month.

The primary value in AI products comes from harnesses - tools that run agents and manage agent fleets - as they route work to the most appropriate model, impose strong guardrails, and manage agent fleets efficiently. Frontier models will continue to play a role in research and development as capital expense, while open-weight models become operational expenses.