Buy a Mac Now or Wait? My Honest Call 8 Days Before Apple’s Oct 28 Event
bufferAbout the author9 min read
Every October I get the same message from friends: should I buy a Mac now or wait? This year the answer actually has a date on it. Apple is holding a launch in Los Angeles on Wednesday, October 28, and if you’ve got a MacBook Pro sitting in your cart right now, I’d close that tab for a week. Not for every Mac, though. I’m writing this on October 20, 2026, before anything has been announced, so a chunk of this is rumor and I’ll label it every time. I’ll update the post after the event.
My angle is the usual one here: help you spend right on AI and hardware. So beyond the buy a Mac now or wait part, there’s a section on how much unified memory you actually need if you want to run local AI models, because that’s where people overspend (or, worse, underspend and regret it in three months).
What’s confirmed and what’s just rumor
Confirmed: Apple’s Greg Joswiak posted “Can’t wait to pro and tell next week” on October 19 (US time), and Apple invited media to a “Pro and tell” experience in Los Angeles on October 28 at 10 a.m. Pacific. That’s 1 p.m. Eastern, 6 p.m. in Madrid and 1 a.m. on October 29 in Beijing. Sources: 9to5Mac and MacRumors. MacRumors expects it to look more like a press-release-plus-hands-on day than a classic keynote.
Rumor, according to Bloomberg’s Mark Gurman: a touchscreen MacBook Pro with an OLED display, an M6 version of the entry 14-inch MacBook Pro, a new iMac and an OLED iPad mini (his product list on X). He also pointed out that the invite art shows a finger drawing on a screen, which is a pretty loud hint about touch (post). 9to5Mac’s summary of his reporting says the OLED MacBook Pro sticks with M5 Pro and M5 Max chips. MacRumors adds that Gurman thinks the new machines could go on sale Friday, November 6. None of that is official.
What I didn’t find: any trustworthy price or configuration leak for the new models. There are numbers floating around, but none with a source I’d put my name next to, so you won’t see them here.
One thing that is official and gets mixed up a lot: M6 already exists. Apple launched it on August 25 in the Mac mini, with up to 32GB of unified memory and up to 170GB/s of bandwidth (Apple Newsroom). The MacBook Pro you can buy today is still M5, M5 Pro or M5 Max. So when people say “wait for the M6 MacBook Pro”, for now that’s about the cheapest Pro only. Keep that in mind for the whole buy a Mac now or wait thing.
Buy a Mac now or wait: my decision table
Here’s how I’d call it, Mac by Mac. Short version: if the product is on Gurman’s list, wait eight days. If it isn’t, waiting buys you nothing except a week without a computer.
| Mac | My call | Why |
|---|---|---|
| 14″ MacBook Pro (M5) | Wait | Rumored M6 replacement at the event. Even if you want the M5 one, you’ll know in 8 days. |
| MacBook Pro M5 Pro / M5 Max | Wait, unless you need it this week | Rumored OLED touch model on top. Per the same rumor the chips stay M5 Pro/Max, so for local AI the old one isn’t worse. |
| iMac | Wait | New iMac rumored for Oct 28. |
| Mac mini | Buy | Just refreshed with M6 on Aug 25 (M5 Pro option too). Not on the rumor list. |
| Mac Studio | Buy | Refreshed Aug 25 with M5 Ultra. Not on the rumor list. |
| MacBook Air / MacBook Neo | Buy if you need one | Not mentioned in this event’s rumors. |
| Deadline next week, any Mac | Buy | A working laptop today beats a better one on Nov 6. |
The MacBook Pro row is the one I keep going back and forth on. If the rumor is right and the OLED model uses the same M5 Pro and M5 Max, then for running models the new one won’t be faster. Same chips, and as far as I can tell nobody has leaked different memory options. You’d be paying for a screen, touch and a lighter body. Some people want exactly that. I’d still wait the eight days, because you’ll find out what Apple does with the current model’s price and stock, and you lose nothing by seeing it.
And no, I wouldn’t count on a price drop. Apple raised Mac prices worldwide in June because of memory costs, and the base 14-inch MacBook Pro went from $1,699 to $1,999 (Reuters). Nothing I’ve read suggests that’s reversing.
How much unified memory for local AI
This is where I’d focus your money, whichever way you go on the buy a Mac now or wait question. On a Mac the GPU uses the same memory pool as everything else, and that pool is fixed the day you buy it. You can’t add RAM later. Ever.
The math is pretty simple once you know three numbers:
- Model file size. On Ollama, a 4-bit (Q4_K_M) 8B model like Llama 3.1 8B is 4.9GB. Qwen3 14B is 9.3GB, Qwen3 32B is 20GB, Llama 3.3 70B is 43GB, gpt-oss-120b is 65GB (Qwen3, Llama 3.3, gpt-oss tag pages). Rule of thumb: at 4-bit, roughly 0.6GB per billion parameters.
- What macOS lets the GPU use. Not all of it. By default it’s about three quarters on bigger machines. A 64GB Mac reports a recommendedMaxWorkingSetSize of 49,152MB, which is 75%, and smaller machines get closer to two thirds (llama.cpp issue #1870). You can raise it with
sudo sysctl iogpu.wired_limit_mb=N, as the MLX-LM README explains, but it resets on reboot and you’re borrowing from the OS. - Room for context. The KV cache grows with your prompt. From Qwen3-32B’s config (64 layers, 8 KV heads, head size 128), I get about 256KB per token at FP16, so roughly 2GB at 8K tokens and 8GB at 32K. Budget 2-3GB for normal chats, more for long documents.
If you’re stuck on buy a Mac now or wait, honestly, this table matters more than the event. Put together, here’s what I’d expect to run comfortably at default settings. “Tight” means it loads but you’ll want short context or a sysctl tweak.
| Unified memory | Available on | Comfortable | Tight |
|---|---|---|---|
| 16GB | M5 (and M6 Mac mini) | 7-9B at 4-bit | 12-14B at 4-bit |
| 24GB | M5, M5 Pro | 14B at 4-bit, 8B at 8-bit | gpt-oss-20b (14GB) |
| 32GB | M5 (and M6 Mac mini) | 27B at 4-bit, gpt-oss-20b | 30-32B at 4-bit |
| 36GB | M5 Max 32-core GPU only | 30-32B at 4-bit | 35B MoE at 4-bit |
| 48GB | M5 Pro, M5 Max 40-core | 30-35B incl. MoE, 27B at 8-bit | 32B at 8-bit |
| 64GB | M5 Pro 20-core GPU, M5 Max 40-core | 32B at 8-bit | 70B at 4-bit (43GB vs ~48GB budget) |
| 128GB | M5 Max 40-core GPU only | 70B at 8-bit, gpt-oss-120b | 100B+ MoE with long context |
A few things that annoyed me while working this out. The 36GB option only exists on the 32-core-GPU M5 Max and you can’t change it, which makes that machine an odd pick for AI: Max money, Pro-level memory. 128GB is only on the 40-core M5 Max. And there’s no 96GB tier at all, so the jump from 64GB to 128GB is one big step.
Bandwidth decides how fast it types
Memory size decides what fits. Bandwidth decides speed. Apple lists 153GB/s for M5, 307GB/s for M5 Pro, 460GB/s for the 32-core M5 Max and 614GB/s for the 40-core one (MacBook Pro specs). In the community llama.cpp benchmark thread, a 7B 4-bit model generates about 32 tokens/s on M5, 68 on M5 Pro and 120 on the 40-core M5 Max (discussion #4167). Pretty much proportional. For the record, the M6 Mac mini’s 16GB config is 153GB/s and only the 24GB and 32GB ones get 170GB/s (Mac mini specs).
What the extra memory costs
Whether you buy a Mac now or wait a week, this is roughly the bill. I checked Apple’s US configurator on October 20. Every memory step costs the same per gigabyte: $25. 16GB to 24GB on the M5 is +$200, 24GB to 48GB on the M5 Pro is +$600, 48GB to 64GB is +$400, and 48GB to 128GB on the 40-core M5 Max is +$2,000 (apple.com). Before June’s price rise, MacSparky reported the 128GB step at +$1,000, about half of today (MacSparky). Ouch.
| Config (US, before tax) | Price |
|---|---|
| 14″ M5, 16GB / 1TB | $1,999 |
| 14″ M5, 24GB / 1TB | $2,199 |
| 14″ M5, 32GB / 1TB | $2,399 |
| 14″ M5 Pro (15-core CPU), 24GB / 1TB | $2,499 |
| 16″ M5 Pro, 48GB / 1TB | $3,599 |
| 14″ M5 Max (32-core GPU), 36GB / 2TB | $4,099 |
| 16″ M5 Max (40-core GPU), 48GB / 2TB | $4,999 |
| 16″ M5 Max (40-core GPU), 128GB / 2TB | $6,999 ($4,999 + $2,000) |
What I’d actually buy
So here’s my buy a Mac now or wait answer by budget. If you only want to try local models next to ChatGPT or Claude, 24GB is the floor I’d accept, and I’d buy it in a Mac mini today rather than wait. 16GB works for 8B models, but you’ll outgrow it the first week.
If local AI is a real part of your work, the 48GB M5 Pro is my sweet spot. It runs the 30B-class models that are actually useful, MoE models included, and it costs $600 more than 24GB, not thousands. Whether that M5 Pro comes in today’s body or the rumored OLED one is a screen decision, not an AI decision.
128GB on the 40-core M5 Max is for people who already know they need 70B at 8-bit or gpt-oss-120b. That’s $2,000 just for memory. Be honest with yourself before clicking that one.
And if the numbers above make you wince, local isn’t the only option. I compared subscriptions and API prices in my AI subscription and API price comparison, and for plenty of people a $20 plan is the cheaper answer to the buy a Mac now or wait question. Already have a Mac that’s suddenly loud and hot? Check whether something is mining on it before you blame its age.
After October 28
I’ll update this post once Apple publishes the new models: real chips, memory options, bandwidth, prices and dates, plus whether the current MacBook Pros stay on sale. Until then, my answer to buy a Mac now or wait is: wait if it’s a MacBook Pro or iMac, buy if it’s anything else and you need it.
Written October 20, 2026 (Asia/Shanghai). Prices from apple.com on that date, US prices exclude sales tax. Rumors attributed to Bloomberg’s Mark Gurman are unconfirmed.
More on this: more AI hardware guides.