Technology news and discussion on TalkFuse.
Open-weight models could cut routine enterprise AI costs while safeguarding data
Gartner analysts Mark Petty and Raymond Paquet argue that open-weight models could handle most routine steps in enterprise AI agent workflows, leaving costly frontier models for tasks that need more advanced reasoning. They say this matters both for reducing costs and for keeping sensitive workflow data within an organisation’s control.
The analysts estimate that nine in ten workflow steps are routine, including classification, tool use and data checks. At the same performance score, open models can cost a quarter as much as the cheapest closed model, or as little as a twentieth of using a frontier model for every step; allowing for extra retries, the practical saving may be five or six times. Gartner’s planning assumption is that open-weight models will run 90% of enterprise operational agent steps by 2030, with software harnesses that route and manage models becoming a key source of value.
In short
- Open-weight models could handle nine in ten routine agent steps.
- Practical savings may be five or six times versus frontier models.
- Gartner expects open models to run 90% of these steps by 2030.
Steelman: both sides, in good faith
The strongest fair case each way — we don't pick a winner.
The case for
Using open-weight models for routine workflow steps could substantially lower costs, especially when frontier models are reserved for the small share of tasks that need stronger reasoning. Running models within an organisation’s own environment may also give it greater control over sensitive data, while model-routing software can match each task to an appropriate system.
The case against
The cost advantage may narrow once retries, integration, maintenance and the software needed to manage models are included, and a cheaper model may not perform reliably enough for every routine step. Keeping models in-house can also shift security and operational responsibilities to the organisation; relying on a mix of models and routing tools may add complexity that offsets some savings and control benefits.
Explained simply
AI agents are systems that carry out sequences of tasks, such as sorting information, checking data and using software tools. In a business workflow, some steps are straightforward while others need more complex reasoning.
Open-weight models make their underlying model weights available for organisations to use and run, including on their own infrastructure. Closed models are controlled by their providers and are often accessed through paid services. Organisations may use more than one model, with software deciding which is suited to each step.
The choice can affect both running costs and where workflow data is processed. It also raises questions about how organisations manage and connect different models while maintaining reliable results.
Read the full article at the source →
Originally published by The Register as “Open models can do 90 percent of agentic work at a twentieth of the cost of a frontier LLM: Gartner”.