Open Gartner: Open Models Could Handle 90 Percent

Gartner’s open-model finding could change how companies budget for AI agents. The firm reportedly says open models may handle 90 percent of agentic work at one-twentieth the cost of a frontier LLM. Open models are AI systems whose weights can be used outside a closed, hosted service.

The estimate reportedly appeared in a Gartner research report published November 3, 2026. That cost gap matters because agentic systems may make many model calls to complete a single business process. Still, the headline figure needs careful reading: it describes a share of workflow steps, not a promise that open models can replace frontier models for every task. Here’s what the estimate could mean for buyers weighing cost against capability.

What does Gartner’s open-model estimate actually measure?

Gartner’s estimate covers agentic workflow steps, rather than claiming that open models match frontier systems across every type of AI task. In their research document, “Open-Weight Models Will Run 90% of Enterprise Agentic Workflow Steps,” analysts Mark Petty and Raymond Paquet describe routine operations such as classification, tool calling, data validation and reconciliation.

These tasks can recur throughout a larger process. That distinction matters. A model might classify records or pass information between tools without having to solve the workflow’s hardest reasoning problem. The Theregister coverage says Gartner sees open-weight models as suitable for much of this routine work. So where might the savings come from?

Why could open models cost one-twentieth as much?

Gartner estimates that open models cost one-twentieth as much to use as a frontier LLM. Put simply, that’s a 20-to-1 cost comparison, not a universal cost guarantee.

Which agent tasks could move to open models?

The examples listed in Gartner’s document point to operational tasks: classifying information, calling tools, passing data between steps, validating fields and reconciling records. But these tasks don’t look the same at every company. A finance workflow, for example, may need stricter checks before a model can approve or update a record than a low-risk internal routing task.

Companies considering open-weight systems should test them with their own documents, tools and failure conditions. Growing interest in Chinese AI startups and models also shows how the open-model market is expanding, though Gartner’s estimate doesn’t name a winning supplier. The key question is whether a model can perform its assigned step accurately enough to avoid costly human corrections. That’s where the harness comes in.

If models become cheaper, what becomes valuable?

A cheaper model alone won’t make an agent dependable. The surrounding software needs to choose the right model, supply relevant context, call tools safely, check outputs and recover when a step fails.

Gartner’s report describes models as increasingly commoditized and directs attention to the systems that organize and control them. That shifts procurement away from simple model-to-model comparisons and toward assessing the entire workflow. Buyers need to compare task accuracy and total operating costs. Bottom line: Gartner’s estimate suggests companies could cut agent costs by routing routine work to open models while reserving frontier LLMs for tasks that justify their higher cost.

Gartner’s estimate: open models could handle 90 percent of agentic work at one-twentieth the cost of a frontier LLM.

What should enterprise buyers verify before switching?

Does the estimate guarantee a 20-fold saving?

No. Gartner’s figure is an estimate, and the available details don’t specify a universal workload or billing basis. Does 90 percent mean every agent can use an open model for nearly all its work? No.

The claim covers workflow steps; it doesn’t show that open models can replace frontier systems for every task. A sensible next step is to identify repeatable, lower-risk work and test models on real company data before changing production workflows. Gartner’s open-model finding makes that review more pressing, but every organization still needs evidence from its own tests. “open gartner” title: “Open Models vs. Frontier LLMs: Gartner’s Insights on Cost Efficiency”