The reviewer installed the open-source Hermes Agent and picked a 125-billion parameter Qwen model, citing privacy concerns.
A Verge reviewer installed local AI agents and large language models to test whether running models on personal hardware beats trusting cloud services. The reviewer used the open-source Hermes Agent on a 256GB M5 Ultra Mac Studio and selected a 125-billion parameter Qwen model that occupies roughly 105GB of storage.
Antonio G. Di Benedetto, a Verge reviewer who covers laptops, wrote that handing personal data to cloud providers held him back from regular AI use. Apple markets its new Mac desktops for local AI workloads, and new RTX Spark Windows machines ship with up to 128GB of RAM aimed at agentic AI.
Local runs remove per-token costs, so builders can experiment with far larger models than cloud budgets normally allow. The tradeoff is hardware: 125-billion parameter models need roughly 105GB of memory, which puts capable local agents out of reach for most standard laptops and desktops.
The reviewer describes the process as overwhelming because the model landscape is large and some models serve specialized purposes. He plans to keep testing Hermes and local models across other Mac and Windows systems, and he will report on whether the privacy benefits justify the hardware cost.
What matters
- The Verge reviewer installed Hermes Agent, a free self-hosted desktop app for macOS, Windows, and Linux.
- Local models of 125 billion parameters can run on machines with 256GB of unified memory.
- The reviewer plans to test local AI across additional Mac and Windows systems over time.
Why it matters
The reviewer plans to test local AI across additional Mac and Windows systems over time.
This GenAI News article was prepared in original wording using reporting and materials published by The Verge. Source reference: https://www.theverge.com/ai-artificial-intelligence/1008484/learning-local-ai-hermes-qwen-model-mac-studio-testing-diary-progress.
Drafted by the GenAI News review pipeline.