MacStories has posted a detailed review of the M5 Ultra Mac Studio, and the verdict is clear: this machine is a major leap for local AI workstations. The reviewer runs personal assistants entirely on a local model, saving real money on cloud subscriptions while keeping privacy intact.
The Winner Against the M3
The M5 Ultra Mac Studio ships with 256 GB of RAM. The reviewer tested it against the M3 Ultra with 512 GB of RAM. For its size, price, and thermal performance, the M5 Ultra wins. Apple’s approach to unified memory helps too.
The RTX 5090 still edges out the M5 Ultra thanks to higher memory bandwidth. But the reviewer prefers the Mac Studio for everyday use. The reviewer notes the PC build is huge, noisy, and hot. Plus, the Mac comes with an operating system that “looks nice and doesn’t suck,” with a vibrant app ecosystem.
The 99-Day Project
The reviewer worked on a major project this summer. They built an internal app called Desk to organize hundreds of notes, sessions, PDFs, and clipped webpages tied to iOS and iPadOS 27. The project grew to 310 documents.
A team of agents ran 24/7 for 99 days. They transcribed WWDC sessions, extracted features from clipped webpages, cross-referenced sources, extracted bugs from screenshots, and worked with the Notion API. The whole setup cost nothing.
Why Local AI Makes Sense
Cloud frontier models are better and often faster. But local AI offers privacy and control. The reviewer says some people use it for sensitive data, others for cool factor, and developers for training. The reviewer’s reasons include the “cool, why not?” factor, privacy, and costs.
The reviewer admits local AI is fiddly. They would never recommend it to someone who just wants to pay $20 for Claude Cowork. This setup is the bleeding edge of AI workflows.
Codex and Qwen3.8-Flash-Next
The reviewer uses local models in the Codex app, orchestrated by GPT-6 Astra. The experience has been great. The reviewer also made the latest Qwen3.8-Flash-Next model their default in Open Minis for iOS and Hermes Agent.
The Numbers Behind the Review
The reviewer ran many tests over four days. The numbers and visualizations come from those tests. macstories.net’s review mixes performance data with practical use cases for local AI applied to workflow.
| Machine | RAM | Comparison Point |
|---|---|---|
| M5 Ultra Mac Studio | 256 GB | Size, price, thermal performance |
| M3 Ultra | 512 GB | Tested against the M5 Ultra |
| Desktop PC (RTX 5090) | Not stated | Higher memory bandwidth |
What Local Models Deliver
Local models start responding more quickly. They stay fast at larger context windows. Long, multi-turn loops run without slowing down.
Real-World Performance
The comparison against the M3 Ultra shows the M5 Ultra’s strength. The tests over four days produced the data that underpins the verdict.
The Bottom Line
The M5 Ultra Mac Studio changes how local AI works. It makes personal assistants fast enough to replace cloud-based ones. The reviewer’s experience proves it.
For anyone curious about local AI, the review shows what the hardware can do. The 99-day project is the proof: local models can handle heavy, always-on tasks without breaking the bank.
Source material: “M5 Ultra Mac Studio Review,” macstories.net.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.