Can a Mac Mini Replace your $200 AI Subscription?
67K views · Sep 28, 2026 · People & Blogs
Comments · 180
@CourtneyRoberts1982 · 1 day ago
Claude wrote the script. It’s got all of the anthropic Hallmarks. That tickles me.
44
@trsd8640 · 2 days ago
well yes, any local llm can't compete with frontier models. But its like cooking yourself: It takes more time, money and work.. and at the end it tastes meh compared with the restaurant. But hey, some of us are having fun with this mess.
29
@Chops00 · 2 days ago
I ordered a M6 Mac mini 32GB initially and ended up canceling it and ordering a M5 Pro 48GB instead. Main reason being Thunderbolt 5 for clustering in the future. But memory bandwidth over the M6 is a huge benefit as well.
34
@duplicate3886 · 4 hours ago
This is the best video so far, a lot of people get played with all the YouTube tutorials and invest on mac mini. Self hosting is not the same as what 10-20 mins video shows.
1
@quadrotonic · 2 days ago (edited)
Excellent video!! There is a quantized qwen 3.8 27b at 12.1gb that apparently works very well.
2
@Gengurke · 1 day ago
Such a good video, keep up the good work. Finally a realistic take on things
1
@kev-ai-87 · 1 day ago
As someone with a Macbook Pro M5 Max 128 running local models for around 12-18hrs/day (mostly Qwen3.8 Flash Next, Qwen3.8 27b and Ornith 1.0/1.5) and also maxing out a Claude Max 20x and an ChatGPT Pro 20x plan, I can tell: you can't replace a subscription with a Mac Mini. I'm thinking about upgrading to a Mac Studio M5 Ultra with 256 GB or more to run GLM 5.3 Flash with around 50-70 tps or smaller worker models with 100-200 tps. This comes almost close to a subscription with the intelligence from 4-12 month ago.
16
@richardward5891 · 1 day ago
Yes, I came up with the same conclusion but you explained the reasons very well. Another option I looked into was running higher performing open models on remote rental servers such as "RunPod". These are interesting cost/benefit calculations and to me are better use cases than local AI with limited memory local hardware. I was thinking about using a pod instead of Sonnet or Luna when my weekly Anthropic/GPT allocation are tight. And the nice thing about the pods is that you use them as you need them - you are not roped into a monthly payment but a daily one - which is fine for overflow needs. There are several workflow configurations and there are drawbacks - like cpu's getting snatched up by the public but it is worth looking into. I ended up NOT doing this because of the newer anthropic and GPT models costing less. I am always spending part of my development time looking at how to route models for different workflows and now sonnet 5.5 can do even more work - as can Luna. I always try to max out my week. It is especially nice when we get resets but I like anthropic's rest where it doesn't rest the day - but that is a topic for a different video ;)
@Icontile1 · 7 hours ago
models in the next year will absolutely change this math quickly, i think get as much ram as you can afford on the m5 pro, then we will see in 12-18 months what models can do for our normal daily tasks.
1
@sheikhazeemsh · 1 day ago
Amazing to the point videos... thanks Kai
2
@415goat · 1 day ago
Amazing video. Really helpful advice.
1
@synovium · 2 days ago
Thanks so much for doing this analysis. I actually commented on your previous video about this very thing, and I was referencing the $20 plan from GPT rather than using a local LLM. Nevertheless, this was a well-thought-out video. Keep up the great work you're doing.
2
Up next
The Current State of Siri AI
SAMTIME · 13K views
How Fast Is Local AI on Your Mac? (Every Chip, M1 to M6
Kai · 7.3K views
ORE Asset System Deep Dive 2026
BlockChain Economics LAB · 2 views
Oracle Silently Started the AI COLLAPSE With Two Words
Finance Bureau · 21K views
Can a 16GB M6 Mac Mini Really Handle Local AI?
Bart Slodyczka · 651 views
The Godfather of AI : Why The Machines Are ALREADY Conscious -Geoffrey Hinton
Ai Reality · 40K views
Don't Buy a Local AI Hardware Until You See This
Kai · 188K views
I Gave Local AI and the Cloud the Exact Same Job
Alex Ziskind · 810K views
I Tried The LAZIEST Way to Make Money With AI
Mark Tilbury · 4.4M views
The New “Transistor” Everyone Is Desperate to Get
Anastasi In Tech · 17K views
The SHOCKING Truth About The Pyramids No One Tells You
The Diary Of A CEO Clips · 4.6M views
The Best Scrabble Player is GENUINELY insane
s0ngbird · 12M views
No wonder it won 37 Emmys
ScreenPuff · 2M views
This Tiny AI Never Stops Learning (mini-AGI)
Better Stack · 820 views
Bessent PANICS As Bond CRISIS Hits WORST MOMENT - Dutch Say BACK OFF USA
House of El · 85K views
Why I’m Returning the M6 Mac mini
Mac Sound Solutions inc. · 129K views
What Happened To Google Gemini?
Ali H. Salem · 423K views
Every Tech Bubble Obeyed the Same Rule. AI Is Next.
Universal Resilience with JT Yu · 532K views
What It Actually Costs to Run Xiaomi MiMo V2.6 Pro & Flash Locally?
Kai · 19K views
Buying a Mac for AI: The Only Guide You Need
Alex Grankin · 138K views
AI Reanalyzed the Pyramids—And the Findings Stunned Everyone
New Discovery · 1M views
Qwen 3.8 27B Just Got 2× Faster on Mac Same Hardware
Kai · 43K views
Oracle’s $288B Lease: The Debt They Can’t Escape
Kai · 9.1K views
Best Mac for Free Local AI — Stop Paying for ChatGPT (2026)
Arthur Winer and AI Master · 64K views
Don't Buy a Mac Mini for Local AI Until You Pick the Model
Amplify Imagination · 8.2K views
POV: You’re an AI Born 9 Seconds Ago
Species | Documenting AGI · 1.6M views