![[Pre Order] Apple Mac Mini M6 12‑Core CPU 12‑Core GPU 16GB RAM 256GB SSD from $1289 Delivered @ Apple Education Store](https://files.ozbargain.com.au/n/93/972693.jpg?h=01e3c6f4)
Great deal for the newly released M6 chip Mac minis released 22 Sep
[Pre Order] Apple Mac Mini M6 12‑Core CPU 12‑Core GPU 16GB RAM 256GB SSD from $1289 Delivered @ Apple Education Store
430
Related Stores
Comments
- 1
- 2
I remember before the price hike, you could get the base model for like $800, and MacBook Pros were like $2500. That said, even at $1289, this is not a bad price in current markets. such a shame though
- 5
This simply isn't sustainable.
- 1
it is what it is…
- 0
base model for like $800
Mac Mini price increases since 2020 have been impressively modest, in my opinion.
2020 M1 base model RRP: $1,100
2026 M6 base model RRP: $1,449
That's a $350 increase after six years of inflation, and despite the global RAM shortage.
For that extra $350, you're getting twice as much RAM, likely much faster RAM, 1.5 × as many CPU cores and GPU cores, substantially faster CPU and GPU performance, a redesigned and much smaller case, more ports including front-facing ports, and various other spec improvements.
- 14
Good point; the hardest one to accept is MacBook Air going from 1400 to almost 2k. I guess wait a few generations and it wouldn't feel too bad given the upgrades
- 3
I bought the base M4 mini directly from Apple last year for A$999.. so that’s a A$450 increase or roughly a 30% price hike in just one year.
They’re great little machines though and I’m sure people will still buy them regardless.
- 7
@chubbychicken: I got the M4 Base for $741, absolutely love it as my media PC connected to the TV.
- 0
@ThithLord: Love my M4 base for daily use, YouTube, Ozbargain, playing iOS games, etc. They are great for the price. But I guess if you want to jump in the car right now, you will just have to buy at around this price now.
- 0
just rrp?
- 1
RRP is $1449, this is edu discount
- 6
They make a big deal about AI performance increases but does anyone care?
- 1
The previous gen M4 Mac Mini was very popular for locally hosted AI. I think this version will be even more popular
- 2
That was mainly due to being a cheap mac with computer use and openclaw with computer use. Not so much for local LLM
- 4
Oh yeah, I guess you can pay for API keys and run a local agent instead. But the purist in me wants local LLMs!
- 3
@3dprintdeals: This is the rub. I considered it but I couldn't drop so much on a refurb.
If you wanted a solid refurb from Apple for this, it'll cost you $1,859.00 for a M4 32GB Ram and a 512GB SSD
I had refurb ping set up in hope for a bit, but I just gave up. Thanks tech bros
- 0
@panda1010101: You can, but your suggestion that the reason for the sudden high popularity was local AI is incorrect.
I made the same incorrect assumption at the time, as Apple silicon is good for that ( or yes, would be if you could afford 64GB).- 0
@netsurfer: depends on the task and model and what you deem acceptable. qwen3.8 on an m4 pro can take over an hour of thinking or it can spend 20 seconds. gemma 4 26B 4AB is decently fast if thats smart enough for you. 16GB of ram wont cut it though.
- 0
Yes for the Mac Studios now with M5 Ultra. Up to 512gb memory in October for local inference will be cheaper and faster than 2x DGX sparks.
Yep, it will cost you four kidneys
- 2
that's ok, i have 4 kids
- 22
Username checks out 😬
- 2
damn, I only have 2- is dialysis still free?
- 0
The 256GB version is almost $16K , 2x DGX sparks 128GB are around the same price ?
- 0
A DGX spark has ~273GB/s of memory bandwidth. This thing has ~1.2TB/s. So should be around 2.2x faster than 2 DGX sparks. Plus way less interconnect overhead.
- 2
@AwesomeAndrew: Thanks!
I am thinking about the 128GB Max versus 96GB Studio versus RTX 5090 32GB.
The Studio has faster bandwidth, but 32GB less ram than the Max. RTX 5090 would probably be the fastest ?- 0
@congo: Depends on what model you want to run. The 3 main ones right now is Qwen3.8 27B and Deepseek V4 Flash and GLM 5.2. RTX5090 would be the fastest in terms of tokens per second, but intelligence matters too.
- 0
@AwesomeAndrew: Thanks, I had a closer look, I think the 96GB Studio Ultra would be the best balance.
- 0
they do if you want to run a localLLM. More bang for your buck them a Nvidia chipset
- 1
tl:dr: Don't pull the trigger because of their AI/neural engine marketing
M1 Max + M4 Max Mac Studio owner here
No, not at all
There's no easy way to tap that Neural Engine in the chip
Neural Engine access via CoreML API, but that api is closed-source (i.e. Gated and locked inside Apple), not open-source nor public access
most of us running local LLMs still falls back on the GPU + CPU- 9
This isn't correct for M5 and newer. MLX models can use the M5 Neural Engine alongside CPU/GPU
https://machinelearning.apple.com/research/exploring-llms-ml…
- 0
The link you found is about the neural accelerator inside the GPU.
Can you please share details of your stack that is tapping the Neural Engine? With mactop log to substantiate?Apple's MLX co-creator delivered the verdict here: https://github.com/ml-explore/mlx/issues/18#issuecomment-284…
(whether Apple was intentional or unintentional,) once again this is another case where marketing sowed confusionWhat the local-LLM communities want to tap is the Neural Engine. It is separate hardware along with the CPU and GPU and media engine.
One of many rationales is to throw more compute capacity to utilize unified memory's bandwidth.There is nearly no LLMs tapping the Neural Engine because its infrastructure requires an unique format: https://github.com/ollama/ollama/issues/3898#issuecomment-20…
- 2
@frkino: When Apple markets their AI/Neural Engine improvements they're not exclusively referring to one specific piece of silicon. They're referring to benefits of all their 'Neural" features in their chips. This includes the Neural Engine in the GPU chip. From the Mac Mini home page under performance: "And a powerful Neural Accelerator in each GPU core speeds up tasks like running on-device AI models"
They also state that it's "Up to 13.5x faster with LLM prompt processing." and cite the test was done with token generation in LMstudio using the features described in the article I linked in my previous comment.
Saying "Don't pull the trigger because of their AI/neural engine marketing" is kinda dumb. It's still an incredibly powerful (and low power) AI workstation thats very price competitive and well supported.
If you were locally hosting AI models with a $1300 budget what would you buy?
PS: if you really care about legacy ANE support why don't you pick up where Eileen left off…
- -1
@stirlo: The Apple Neural Engine (ANE) is a very specific piece of technology that is different to the "powerful Neural Accelerator in each GPU" they talk about there. That "Neural Accelerator" is the equivalent of what NVidia calls a tensor core.
But the ANE isn't useful for running LLMs. The new Neural Accelerator GPU cores are.
However, even without that the M-series of chips is a good solution for running local LLMs. You do need to understand the trade-offs vs a NVidia system though.
- 0
Apple's MLX co-creator delivered the verdict here: https://github.com/ml-explore/mlx/issues/18#issuecomment-284…
That's him saying they won't support it because it is slower and unsupported (ie, will change per model), not because they can't.
The ANE doesn't have a documented API but people have reverse engineered it, eg: https://maderix.substack.com/p/inside-the-m4-apple-neural-en…
The TL;DR is that is isn't really designed for running LLMs and instead is designed for low power consumption for things like visual models (eg, object tracking for Center Stage feature used in the webcam)
OTOH the M-series of CPU/GPUs is well suited for running LLMs. It is excellent at the decode phase (generating tokens) because of the massive bandwidth of the unified memory architecture.
But it is a lot slower than NVidia GPUs for prompt processing because the raw processing power of the GPU is less (and no, the ANE can't help here). This is knows as the "Prefill phase" and Googling for "Prefill on apple chips slow" will find you lots of good explanations.
What does help with this is the tensor cores in M5 and above. Still not as fast as NVidia cores but the trade off with that memory bandwidth makes it more competitive overall.
most of us running local LLMs still falls back on the GPU + CPU
Well that's what people do when using NVidia too, so not bad. NVidia GB100 is literally a CPU and GPU combination.
- 2
What machine do you recommend getting for LLM?
- 0
how much money do you have?
- 3
Probably an RTX 5090 based system if you really want to run local LLM all the time.
Geekbench 6 AI benchmarks shows the RTX 5090 outperforming the M4 Max across single-precision (+100%), half-precision (+154%), and quantized operations (+46%).
If you don't care about speed and really want to chase after running large LLMs locally, you could get M5 Max or Ultra I guess. Honestly, if you currently not paying and using ChatGPT Pro or Claude Max, do you really need to run local LLM? Even if you do run one, if you really do use a lot of AI, you are still much better off paying for GPT or Claude (much better performance and result). If you are not tech savvy, you probably would use a ChatGPT or Claude agent to help you set it up and troubleshoot.
Most people should be fine with ChatGPT Plus. It's more important to use AI to get things done, rather than mucking around with running local LLMs.
- 1
@netsurfer: I run local on 16GB M4 MBA with Qwen 3.5/9b, it's kinda fun for non-work discussions and a learning tool. I asked it about taxonomy of dinosaurs, for example, hard for a novice to figure out the "big picture" from reading wikipedia. It's slow but still faster than it takes me to read the responses. For work though, when GitHub jacked their prices I subscribed to Claude Pro and rarely hit a limit, using a mix of Haiku and Opus. Getting the same capability locally would never ever have paid off.
- 0
@netsurfer: Yeah I have a few Pro subs and they are great. But I don't want to be paying an expensive "subscription" for life.
- 0
So the M 1-4 had a NPU (used for fingerprint as apple dont use faceID in 2026) so its SLOW as it didn't need to be fast.
The M5 brought the NPU into the GPU cores (45 tops, again its face ID and faster PREFILL for Ai which helps with optimized MLX format Ai)
M6 brings dual NPU cores in the GPU and thats NOT 2x45 tops=90tops but it allows for even faster prefill with the 10% faster memory speeds which is the real bottleneck.The M6 is a laptop chip (hence the focus on lower powered cores) and you need to be looking at the STUDIO M5 MAX&ULTRA 600mbs/1.2tb memory transfer speeds) as thats the real bottleneck for Ai LLM's. Dont buy the M6 thinking you getting an ai chip improvement! Its a LAPTOP chip thats getting a testing rollout before the macbook pro/air model early in 2027… 16gb/256gb is too small to be at all useful too, so factor in that too, 32gb/1tb was 'enough' in 2024/25 its a bad joke in 2026 and has no longevity, just apple profitability!!!
- 0
Yes..
- 0
Yea have a look on youtube, all the AI youtubers are going nuts about how good a deal this is for running local models.
- 0
only 256GB….
- 9
yeah, the classic Apple bait.
- 5
Thanks. But we know how to read the title
- -8
Yet punctuating a sentence still eludes us.
- 1
yep
- 0
The fact that OP says “great deal” makes me want to laugh and cry.
- 8
Possibly user replaceable like the last one, also can buy an external drive/hub. This is desktop land so no need to worry about portability.
- 0
And only 16GB Ram
- 1
I'm surprised it's not 8GB!
- 1
I'd rather have the option of 256GB than not…
- 2
20% increase in multi core. They are hiding the single core performance though
- 6
Wow, what a price increase!
I bought my Mac mini with the M4 Pro chip, 48GB memory and 512GB SSD back in July last year for $2,235 via AOC and Coles gift card
no ragrets.
- 1
Cool story
- -5
Got the same config this July. Was worried I was buying just before the new ones come out but realised I made the right choice.
- 0
wow based model used to be ~$800 now ~$1,2k
- 6
Yes, RAM/SSD crisis are giving Apple more excuse to jack up the price, no thanks.
- -7
is it an excuse if the price of RAM has actually gone up? thats the reality.
- 4
Yes because it hasn't gone up by that much
- -3
Blame RAM manufacturers for price fixing/cartel-like behaviour pouring fuel on the fire on top of shortage. There were even reports from Valve that when RAM manufacturers give you a price and you say anything other than 'yes', some of them don't even bother talking to you again and you have to beg them to buy whatever they originally quoted.
- 0
Blame AI companies for simply throwing exorbitant money around in hopes to become a dominant player in the AI race.
Unfortunately these companies can simply outbid any consumers with their fat wallets.
It's so lucrative that we saw Micron give everyone the finger and say they're chasing that AI datacentre money exclusively.- 1
and I just bought a certified refurbished M5 Pro MBP 14 lol
my CPU isn't the latest anymore (cry)
- 1
the good news is that your CPU is still totally fine?
- 0
my thoughts exactly. I think that's the best choice at the moment for longevity and support, performance too. I don't need Max.
- 0
I bought a 256gb M4 mac mini on launch for $849. I thought I would have upgraded it by now but the replacement cost is outrageous. I should at least be glad that I upgraded when I did.
- 1
I mean, if you don't need it, and can use old hardware/pc then keep it, don't give your hard earn cash to greedy company.
- 0
Have had the M4 16/512GB mini and Studio Display for 8 months in my home office, bought off AOC store, great setup.
Would be a lot more now.- 0
Ordered mine before the AOC store closure and price hikes, yay impulse buying.
- 0
The bare dare that Apple refuses to consume the supply cost and delivery it to us in full. What a joke that 16/256 for $1200+ while marketing it, what, “AI ready”?
- 2
Sure $800 for an M4 Mac mini was a good deal last year but… this is still a good deal. Great little powerhouse.
- -1
Wow an extra $570 for an extra 16GB of RAM? That's hectic, even at current ram prices.
- 2
Apple has always charged a premium for ram a d storage so not too surprising
- 1
so like with the ram tax these days these price are resonable? lol
- 0
No thanks, my old mini can last another three years.
- 0
how to get around without uniday account?
- 4
Why do you need uniday? Just check out through education store
- 0
it requires verification through unidays.
- 0
That's new.
- 0
@3dprintdeals: Agree! Miss the days where you could just buy through the educational store… probably would've pulled the trigger on this if so!
- 0
screw the mini!! who's lining up with me for a new polishing cloth??? https://www.apple.com/au/shop/product/mlh74fe/a/polishing-cl…
- 2
LOL
- 0
A few years from now, I’m going to have a Mac Studio M6 sitting in my homelab.
You won’t notice it. You won’t hear it.
But behind that rack will be models large, retarded AI models running day and night on hardware that was never meant to be used like this.
I don’t know what they’ll generate.
But I have a very particular collection of unhinged AI Models.
And I intend to find out.
- -3
why would apple do this, jump 2 gens, so next year what is it gonna be m7? and m6 pro for laptops?
- 0
idk but to me feels like M1 was a new release just a few years ago.
can't believe already at M6. like chips growing so old so quickly.I'm convinced this is a weird situation for timing. I think for the M6 capability (like some saying can run local AI) it's not too expensive looking. but probably in a few months we will be able to do everything much faster and cheaper cloud based. no more having a PC at home for powerful video editing, no more worrying about 50TB storages and backups for that. it's all going to be cloud based at a cheap price (othewise people will keep buying computers).
massive shortage now on computer parts and price rises for evevrything including router modems because they use chips and memory storage inside. I heard people are buying out SSD drive at current prices, because major manufacturers of memory already announced, they are stop making consumer line memory (price isn't going to drop, and never going to be restocked). it's either buy now at current price or never (according to many recent youtube I watched).
- 0
it cant run local ai better than an m5 pro laptop or mini (which u can configure it now with an m5 pro), m6 base its weaker going by the numbers and its ram limitation atm is 32gb config, its only an upgrade to the m4, also if ur not tapped in with the local llm world, these things can never run a proper local ai in good speeds, if u took the big daddy macbook pro 14/16 with a m5 max then we are talking usable stuff, anything else ur better off using cheap Chinese api this would barely run gemma 4 at 12b 4bit quant.. maybe 20tokens per second at start of context window
- 0
yeah my point is cloud will quickly become cheaper and way faster.
soon I think computer prices will go up even more, making their value less and less (as you said no compete with cloud ai).
I think that's where we are going. to use a terrible example similar to Chromebook situation. nothing installed anymore everything accessed on cloud. what you will have and basically just need to be a laptop with screen and small amount of memory.
- 0
@skibidigrimacmoggr67: You think once everyone is locked into subscription pricing they're going to keep it super affordable?
- 2
@Daff: once they move everyone away from buying PC and running local ai yes I think it will become lower in price so there is no thinking of going the "own PC" way.
obviously after 10~15 years of sub, having your own ai PC might be cheaper, but to keep fast "what comes next" text ai it will be running 10x crappier ai and nobody thinks that far ahead. also I think no hardware going to be left for "cheap" ai PC option in the future.
- 0
Easy, if Apple didn't jump 2 gens, M5 Mini at that price point would get competition from Macbook Air M5 (also there are probably already many people with M5 Macbook Air, most of them purchased before the price hike, so we are looking at people who got Macbook Air M5 at roughly the same price as Mac Mini M6).
Mac Mini M6 needs to have its run first before Macbook Air M6. There is also some speculation that Apple might wait for M7 Pro / Max for next iteration of Macbook Pros (skipping M6 Pro / Max). We don't know how expensive the first gen 2nm manufacturing is (and whether the second gen has cost reduction benefits).
- 0
nah man i got a macbook air m5 before price rise from costco deal price match 15 inch for $1800 (best price since launch) now its $2499 or $2340 edu price, there aint no mac selling for 1300ish dollars, this woulda been the better buy if one was after a desktop setup. laptop is laptop and has its best use cases. this would be ok for a openclaw hermes local 24/7 desktop setup with a api. but then again ur better off paying for a monthly sub vps if your only buying it for openclaw/hermes
- 0
I was referring to 13 inch Macbook Air and the 512GB SSD base model. The current base model of M6 Mac Mini starts with 256GB SSD. Can't remember the best price for Back to Uni offer earlier this year (with free AirPods 4). You really think $1289 for a Mac Mini M5 would be attractive? Even Apple isn't confident on that so it tried to entice people with a CPU with fastest single core performance, up to 20% boost in multi core and more in GPU and AI.
M4 Mac Mini was $815.10. Let's face it openclaw and/or hermes on a $815 Mac Mini is something we can numb our brain on. There were cheaper ways to run openclaw before the RAM and SSD hike. It's mainly just those youTube videos and articles which cause people wanting to run it on a Mac Mini.
Apple picked M6 for the new Mac Mini…. so what? No big deal. There will be M7, M8….
- 0
@netsurfer: oh yes i remember the mini at 815 but if u woulda used the other link which was a bigger discount it was 700! lol i remember contemplating buying it at the time just to learn mac os but didnt. your talking about the m5 13" with like 3 coupon stacks, yeh it was a good deal but not a normal price, and the price u mention is after selling the airpods and considering it 200 worth of coin
- 0
@Roe Jogan: There was also a cashback stacking deal for Mac Mini M4, but it was only for a few days. Even at $815, most OZBers would have used the 10% off Apple cards from Woolies / Coles trick, but such method can still be used for Mac Mini M6.
I think it is just Tim Cook once again showing his mastery at inventory management. Used Mac Mini M4 to clear all the M4 chips. Kept M5 chips for Macbook Airs / Pros and M5 Pro chips for Macbook Pros for as long as possible, taking advantage of the RAM and SSD situation where PC / Windows laptop makers had no answer.
Apple (Tim Cook) could have released M5 Mac Mini last year or early this year so that we could get one for $815. I know many people wanting / hoping for a M5 Pro Mac Mini at a cheap price early this year, but selling more Macbooks is better for Apple's bottom line.
Apple delayed M6 Mac Mini's and M5 Pro Mac Mini's as long as they could. Non tech savvy people can get tricked with those running openclaw / hermes on Mac Mini youTube videos. However, a lot of those youTubers do use paid AI models (if they are doing AI related youTube videos, they would already used them for other AI tasks), but it is not obvious to general public. You can't expect LPDDR5X to do magics and outperform HBM in AI. Buying Macs with large amount of RAM to run mostly local LLMs (and pay a hefty Apple tax) is just silly.
- 0
Holy smokes that’s terrible value
- 2
- 1
- 2
shame about the massive price hikes due to rampocalypse.