Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
The hate for a touchscreen MacBook is unbelievable.It's a third input device - great for scrolling, pinch to zoom, and quick touch targets. It IS NOT replacing you preferred trackpad and keyboard. AND, if you want to run iPad apps on the laptop, it is nearly essential.

BETTER STILL, however, would be an iPad Pro 13' SKU that can run MacOS - don't buy it if you don't need it but let the rest of us have the ultimate portable device for OUR NEEDS.
 
  • Like
Reactions: Tagbert
Everyone thinks the display will be the touchscreen. I think it will be the keyboard and display.
 
If you were leveraging local LLMs to accelerate your development, then you'd both be delivering quicker and could absolutely justify a more recent M-series chip.
I've been trying local LLMs on my M5 Pro today. My conclusion (& you're welcome to disagree) is that it doesn't have anything like enough power to be actually useful for development.
 
I'll wait for the M7 Max MBP. Disappointed they aren't keeping up with their chip release cadence, but I might not be the target audience for a touchscreen/OLED MBP. If there isn't a spec bump I'm not interested. The iPad Pro's screen is genuinely amazing, but not worth buying an entire new MBP to put an iPad Pro's screen on the M5 Max MBP I already own.
 
I've been trying local LLMs on my M5 Pro today. My conclusion (& you're welcome to disagree) is that it doesn't have anything like enough power to be actually useful for development.
I agree with this. Fable 5 and gpt-5.6-sol are truly game changing, and even last-gen LLMs like Opus 4.8 and gpt-5.5, but there is nothing you can run on a single Mac - even the max memory variants of the Studio - that even approaches that level of intelligence.

I'm not saying high-end open weight LLMs don't exist. They do. But there isn't enough RAM to run them on any MBP, even if they bump the RAM to 192GB.

We would need something like 768-1024GB of unified memory on a Mac to be able to run frontier-class open weight models at appreciable speed and accuracy. Those quantizations and distillations might work for extremely trivial examples, but for real work you need the full model and minimal to no quantization.
 
  • Like
Reactions: wyrdness
We would need something like 768-1024GB of unified memory on a Mac to be able to run frontier-class open weight models at appreciable speed and accuracy. Those quantizations and distillations might work for extremely trivial examples, but for real work you need the full model and minimal to no quantization.
The M5 Ultra Mac Studio will be exactly this. Which is why they will cost $30K.

Apple will sell every one they can make, and new-in-the-box models will go for $50K on eBay.
 
The M5 Ultra Mac Studio will be exactly this. Which is why they will cost $30K.

Apple will sell every one they can make, and new-in-the-box models will go for $50K on eBay.
While I completely agree that those machines will sell for $30-50k new, I don't understand why anyone would ever buy one at those prices.

Let's assume that by canceling your cloud LLM subscriptions, you are saving $500/month. That's just a round number to represent the type of subscription prices out there: OpenAI and Anthropic are up to $200/month, SuperGrok is $300/month, and Google's AI Ultra plan is $200/month. If you have 2 plans that's $400/month. So just for the sake of argument let's say $500.

In that case, if you buy a $30k machine tax-free, it will take 60 months -- 5 years -- for the Mac to get out of the investment hole and start returning monetary value back to you compared to your $500 of LLM subscriptions.

I can think of much better ways to invest $30k to get a bigger payback in 5 years. And you also have to consider that you don't have access to that money during the 5 years after you spent it. Many investment opportunities exist where you can pull your money out any time you need it.

Meanwhile, even if the M5 Ultra Mac Studio has 1 TB of RAM, that won't be very much compared to the demands of the frontier models in 5 years. It'll be a lot, but not anywhere close to what the state of the art is capable of using. We should see state of the art models demanding in the 2 to 5 TB range in 5 years. Maybe even more.

So you are locking yourself into 2026 hardware for a huge up-front cost, and the only way you ever see any financial benefit is if you keep the machine longer than 5 years.

Again - I know people will buy this. But tell me how this makes sense.

Edit: I also realized that you will have a difficult time running multiple models in parallel with your purchase of a single 1 TB Mac Studio. Even if you bought two of these beasts, you're looking at just under 2 TB of usable VRAM. If each model takes, say, 800 GB, that means you can run two in parallel.

Meanwhile, a real workflow in Claude Code or Codex can use numerous LLMs in parallel. For some of the more demanding complicated knowledge work and coding, you might benefit from 5-10 LLMs running in separate contexts at the same time. This is not only to save time, but to enable agentic workflows like in Ultracode mode.

Those agentic workflows are really smart because they have parallel researcher LLMs, implementation LLMs, and verifier LLMs that form a directed graph of LLMs working a "flow chart" of orderly and utterly methodical implementation until everything is working 100%. With a huge $60k investment in two 1 TB Macs, you're still limited to a fraction of the productivity of someone paying $200/month for Claude Code (or $400 for two subscriptions).

It really doesn't make financial sense at all to me.
 
Last edited:
  • Like
Reactions: wyrdness
Then don't touch it. Or wash your damned hands. Or just blow your nose instead of jamming your fingers up in there!
Most people have these things in their bodies called "pores" and "glands". Even with regular hand washing and good hygiene habits, their skin and fingers have oil on them. Yours does too. It's nothing to be ashamed of.

Not only does "not everything has to be for everyone" ring 100% true, it's a fact that it is "the rich" (or at least people willing to work hard enough) to come up with the money and help fund future R&D by purchasing all those newfangled things that the 98% reject.
"People willing to work hard enough"?? You sound like a character from a Charles Dickens novel. Are you saying that poor and economically disadvantaged people have only themselves to blame because they're lazy?

News flash, Ebenezer: Being able to afford something isn't the same as being willing to pay its price.
 
While I completely agree that those machines will sell for $30-50k new, I don't understand why anyone would ever buy one at those prices.

Let's assume that by canceling your cloud LLM subscriptions, you are saving $500/month. That's just a round number to represent the type of subscription prices out there: OpenAI and Anthropic are up to $200/month, SuperGrok is $300/month, and Google's AI Ultra plan is $200/month. If you have 2 plans that's $400/month. So just for the sake of argument let's say $500.

In that case, if you buy a $30k machine tax-free, it will take 60 months -- 5 years -- for the Mac to get out of the investment hole and start returning monetary value back to you compared to your $500 of LLM subscriptions.
Those are hobbyist numbers. Make them 10-25x for proper business/institutional/corporate use and the machine pays for itself in six months. Six weeks in some cases.

Plus you are paying for privacy. Which is mission critical for many orgs, even if the Mac Studio costs more or is less efficient than a cloud subscription.
 
I'm still using my USB-A external BluRay drive, SSD enclosures and harddrives, memory card reader, and webcam (my MBP is in clamshell mode 95% of the time). They're all connected to my CalDigit dock.

I'm not saying the MB Ultra needs to include USB-A ports (because new stuff is predominantly USBc), but there's still plenty of need and uses for the legacy port.
I was joking. surely that was obvious from my tone?
Personally, I still buy CDs and use my USB-A external drive to rip them to my Mac. With an adapter, of course. Because USB-A is not allowed.
 
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.