# M5 Ultra Mac Studio testing shows faster local AI agent workloads

_Published Monday, September 21, 2026 at 12:19 PM EDT · AI, Infrastructure · Latest · Tier 2 — Notable_

![M5 Ultra Mac Studio testing shows faster local AI agent workloads — Primary](https://cdn.macstories.net/images/uploads/2026/09/21/shortcut-upload-1789994617403-abb772eab6.png)

MacStories' hands-on testing found that an M5 Ultra Mac Studio processed prompts about 2.5 times as fast as its M3 Ultra predecessor and generated responses about 70% faster on average in tested local-model workflows.

The machine ran Qwen3.8-Flash-Next above 100 tokens per second on short prompts and at 60 to 85 tokens per second with 64K to 256K of context. Its 1.2 TB/s memory bandwidth helped sustain multi-turn agent sessions, although an RTX 5090 retained a bandwidth advantage and cloud models remained more capable and potentially cheaper than buying the hardware.

## Sources

- [macstories.net](https://www.macstories.net/stories/m5-ultra-mac-studio-review-the-dream-mac-for-local-ai-agents/)

---
Canonical: https://techandbusiness.org/newswire/a8DniYMfvsclAgQgNRw-7c
Published: 2026-09-21T16:19:49.451Z
Story chronology: 2026-09-21T13:00:02.000Z
Retrieved: 2026-09-21T18:20:07.679Z
Publisher: Tech & Business (techandbusiness.org)