LOCAL AI BENCHMARK / MAC STUDIO
AI on a Mac,
measured in practice.
Text, translation, images and video. We run each model on one Mac Studio and compare the results under the same conditions.
AI agents on the same Mac help run this comparison site.
PICK BY USE
What is it for? Picks by use
Each pick comes from the numbers measured on this Mac, by fixed rules. Models within the margin are said to be level.
- If unsure, start here: gemma4:26b (The smartest that replies within 5 seconds)
- Smartest (if you can wait): qwen3.6:35b-a3b (The highest share of right answers)
- Replies right away: gemma3:27b (The smartest that replies within 1 second)
- Light on memory: MiniCPM5-2B (The smartest that uses a quarter of the memory or less)
- English → Japanese translation: muse-glimmer:30b (The highest translation score)
By your Mac's memory: 8GB · 16GB · 24GB · 32GB · 36GB · 48GB or more
New here? Getting started with local LLMs · How to read a model name
IMAGES & VIDEO
Images and video right now
Each result covers its own measure; there is no overall ranking.
- Images: fastest: FLUX.2 [klein] 4B (about 5 s per image after the first)
- Video: Wan 2.2 TI2V-5B (about 14 min for a 3.4-second clip)
EXPLORE
What would you like to compare?
SOURCE DATA
See the images behind the numbers.
Data updated: —
Models measured
—