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

Gartner predicts that open-weight models will handle 90% of enterprise agentic AI workflow steps at a fraction of the cost compared to frontier large language models. The firm's analysts, Mark Petty and Raymond Paquet, claim that routine steps in agent workflows, such as classification, tool calling, passing, validating, and reconciling data, can be performed by open-weight models with parity to the cheapest closed models at a quarter of the price.

While current open models may require more resources, the savings per task against a frontier model are estimated to be around five or six times. By 2030, the analysts project that 90% of enterprise operational agent workflow steps will execute on open-weight models. Enterprises will primarily adopt domain-specific open-weight models for scenarios with a small number of highly stable steps run millions of times a month.

Harnesses, which manage agents and route work to the most appropriate model, are considered the source of value in AI products. Frontier models will continue to play a role in research and development, while open-weight models become an operational expense in the heavy lifting.