Prices for all 9 Mac mini M6 & M5 Pro configurations (VAT included)
Mac mini M6 starts at 27,780,000₫, M5 Pro at 53,340,000₫. Each card shows the best AI model it can run with its AA score — filter by chip, RAM or the model you need.
Compare configurationsCompare the Mac mini chips side by side, or tick “Compare” on any card (up to 4 machines, Mac Studio and Mac mini can be mixed).
Prices include VAT. SSD upgrades and 10Gb Ethernet cost extra — choose them on each machine’s detail page. Specifications as published by Apple, updated September 24, 2026.
Compare 9 configurations at a glance: RAM, bandwidth and AI model
| Configuration | CPU / GPU | Memory | Bandwidth | Best model (AA score) | Price |
|---|---|---|---|---|---|
| M6 12/12 · 16GB | 12 / 12 cores | 16GB | 153GB/s | gpt-oss-20b · no AA v5 score yet | 27,780,000₫ |
| M6 12/12 · 24GB | 12 / 12 cores | 24GB | 170GB/s | Qwen3.8-27B (4-bit, tight on memory) · AA score 20.1 | 33,900,000₫ |
| M6 12/12 · 32GB | 12 / 12 cores | 32GB | 170GB/s | Qwen3.8-27B (4-bit, tight on memory) · AA score 20.1 | 40,010,000₫ |
| M5 Pro 15/16 · 24GB | 15 / 16 cores | 24GB | 307GB/s | Qwen3.8-27B (4-bit, tight on memory) · AA score 20.1 | 53,340,000₫ |
| M5 Pro 15/16 · 48GB | 15 / 16 cores | 48GB | 307GB/s | Qwen3.8-27B · AA score 20.1 | 71,670,000₫ |
| M5 Pro 15/16 · 64GB | 15 / 16 cores | 64GB | 307GB/s | Qwen3.8-27B (8-bit) · AA score 20.1 | 83,900,000₫ |
| M5 Pro 18/20 · 24GB | 18 / 20 cores | 24GB | 307GB/s | Qwen3.8-27B (4-bit, tight on memory) · AA score 20.1 | 59,450,000₫ |
| M5 Pro 18/20 · 48GB | 18 / 20 cores | 48GB | 307GB/s | Qwen3.8-27B · AA score 20.1 | 77,780,000₫ |
| M5 Pro 18/20 · 64GB | 18 / 20 cores | 64GB | 307GB/s | Qwen3.8-27B (8-bit) · AA score 20.1 | 90,000,000₫ |
Memory bandwidth determines how fast the machine generates text. GPU core count determines how quickly it processes a long prompt before it starts answering.
How much RAM do you need for which AI model?
How much RAM you need to run an LLM locally on Mac mini depends on the model you need: the 16GB version runs gpt-oss-20b · no AA v5 score yet at best, the 64GB version runs Qwen3.8-27B (8-bit) · AA score 20.1. Memory capacity decides which model fits, the AA score shows how smart that model is, and memory bandwidth decides how fast it runs. The table below lists the best model each memory tier can run. Mac mini suits testing or personal use, not as the AI server of a Namtech internal-AI package.
| System memory | Available to the model | Best model it can run (AA score) | Best for |
|---|---|---|---|
| 16GB | ~12GB | gpt-oss-20b · no AA v5 score yet | Local AI trials, summarizing and drafting short documents |
| 24GB | ~18GB | Qwen3.8-27B (4-bit, tight on memory) · AA score 20.1 | Personal use; Qwen3.8-27B runs but tight on memory |
| 32GB | ~24GB | Qwen3.8-27B (4-bit, tight on memory) · AA score 20.1 | As above, with extra headroom for longer context |
| 48GB | ~36GB | Qwen3.8-27B · AA score 20.1 | Personal use, trying document Q&A with longer context |
| 64GB | ~48GB | Qwen3.8-27B (8-bit) · AA score 20.1 | The 8-bit version — the highest memory tier of Mac mini |
Memory available to the model ≈ 75% of RAM (the macOS default for graphics). AA score: Artificial Analysis Intelligence Index v5.0, checked 3 Nov 2026 — higher means smarter; the v5 scale is harder, so scores are lower than before. For comparison, Claude Opus 5.5 scores 58.4 and the leader, Gemini 4 Carbon, scores 59.4. Response speed and the number of simultaneous users depend on the GPU and memory bandwidth; Namtech tests on real hardware before advising.
Mac mini suits testing or personal use, not as the AI server of a Namtech internal-AI package.
Mac mini M6 or M5 Pro: which one for local AI?
Choose M6 when you are just starting to trial local AI: the 32GB version runs Qwen3.8-27B (4-bit, tight on memory) · AA score 20.1 at best. Choose M5 Pro when you want more memory headroom (the 64GB version runs Qwen3.8-27B (8-bit) · AA score 20.1): it offers 24–64GB of memory and 307GB/s of memory bandwidth, versus 153GB/s – 170GB/s on M6.
| M6 | M5 Pro | |
|---|---|---|
| CPU / GPU | 12 / 12 cores | 15/18 / 16/20 cores |
| Neural Engine | Dual 16-core Neural Engine | 16 cores |
| Unified memory | 16–32GB | 24–64GB |
| Memory bandwidth | 153GB/s – 170GB/s | 307GB/s |
| Thunderbolt ports (back) | 3× Thunderbolt 4 (USB-C, up to 40Gb/s) | 3× Thunderbolt 5 (USB-C, up to 120Gb/s) |
| Storage | 256GB – 2TB | 512GB – 8TB |
| Best model it can run (AA score) | Qwen3.8-27B (4-bit, tight on memory) · AA score 20.1 | Qwen3.8-27B (8-bit) · AA score 20.1 |
| Price (VAT included) | 27,780,000₫ – 40,010,000₫ | 53,340,000₫ – 90,000,000₫ |
Bandwidth determines how fast text is generated; GPU core count determines how quickly a long prompt is processed. Chip, RAM and bandwidth figures as published by Apple, checked September 24, 2026; bandwidth for the M5 Pro with 18-core CPU and 20-core GPU comes from Apple’s MacBook Pro spec page, as the Mac mini page does not list it separately.
When Mac mini is enough, and when you need Mac Studio
Mac mini is enough when
You are just starting to trial local AI or using it personally, mostly summarizing and drafting short documents. Mac mini offers 16–64GB of memory. Mac mini suits testing or personal use, not as the AI server of a Namtech internal-AI package.
Move up to Mac Studio when
You need an AI server for a whole team, a smarter model, long context, many people asking at once, or several models on one machine. Mac Studio offers 36–512GB of memory (the 512GB version runs GLM-5.3-Flash · AA score 27.8).
The lowest-priced Mac Studio configuration is the Mac Studio M5 Max 18CPU 32GPU 36GB 512GB at 77,780,000₫, whose best model is Qwen3.8-27B · AA score 20.1. Our advice is straightforward: if a Mac mini is already enough for your workload, Namtech will tell you before you spend money on a Mac Studio.
On-premise AI on Mac mini: your data never leaves the office
On-premise AI means the model runs on a machine located at your company. Questions, documents and answers all stay on your internal network — the foundation for an internal chatbot, a Q&A assistant over your document library, file summarization and drafting.
Unified memory
The CPU and GPU share a single pool of memory. The whole model loads onto the machine, with no need to split it up across several graphics cards.
Fits in the office
Height 5.0 cm · Width 12.7 cm · Depth 12.7 cm. Sits neatly on a desk, with no need for a server room or rack cabinet.
Low power draw
Maximum continuous power of 155W for the whole machine, per Apple’s specifications.
Your data stays in-house
The model runs locally and does not call any outside service. Suitable for HR records, contracts, customer data and internal documents.
Genuine machines and accessories are sold and VAT-invoiced by Công ty Cổ phần Trạm Năng Lượng Số; Namtech advises on the configuration and can install and tune the software that runs your models (MLX, Ollama or LM Studio) to your requirements (quoted separately).
Buying from Namtech vs buying directly from Apple Store
| Buy from Namtech | Buy directly from Apple Store | |
|---|---|---|
| Payment | Only a 20% deposit when the order is confirmed; the remaining 80% is paid before delivery | Full payment when ordering, even if you wait weeks for stock |
| Cancellation | Cancel before the order is confirmed: free of charge | Cancel once the order is being prepared for shipping: 20% restocking fee |
| AI advice and setup | Configuration advice based on the AI models you plan to run, help installing MLX, Ollama, LM Studio | Not offered |
| Delivery | Delivered to your door, factory-sealed; you open the seal yourself | Free delivery |
| Invoice | VAT invoice issued by Công ty Cổ phần Trạm Năng Lượng Số | Apple invoice |
| Warranty and returns | Per Apple’s policy | Per Apple’s policy |
Specifications common to all configurations
| Ports | Back: 3× Thunderbolt 4 (M6, up to 40Gb/s) or 3× Thunderbolt 5 (M5 Pro, up to 120Gb/s) — USB-C, USB 4 up to 40Gb/s, DisplayPort · 1× HDMI (multichannel audio support) · 1× 2.5Gb Ethernet (RJ-45) — configurable to 10Gb Ethernet (Nbase-T) · Front: 2× USB-C (USB 3, up to 10Gb/s) + 3.5 mm headphone jack · Wi-Fi 7 (802.11be), Bluetooth 6, Thread — Apple N1 chip |
|---|---|
| Display support | Up to 3 external displays · M6: two 6K@60Hz displays + one 5K@60Hz display · M5 Pro: three 6K@60Hz displays · with up to two displays: one display up to 8K@60Hz |
| Dimensions | Height 5.0 cm · Width 12.7 cm · Depth 12.7 cm |
| Weight | 0.67 kg or 0.73 kg depending on chip (per Apple’s spec page) |
| Power | 100–240V AC, maximum continuous power 155W · operating temperature 10–35°C |
| In the box | Mac mini and power cord (1.8 m). Display, keyboard and mouse sold separately. |
Source: Apple’s official technical specifications, checked on September 24, 2026.
Frequently asked questions about Mac mini for local AI
Can Mac mini run local AI?
Yes, for testing or personal use. The lowest-priced version (16GB, 27,780,000₫) runs gpt-oss-20b · no AA v5 score yet at best; the 64GB version runs Qwen3.8-27B (8-bit) · AA score 20.1. Mac mini suits testing or personal use, not as the AI server of a Namtech internal-AI package. If you need an AI server for a team or many people asking at once, choose Mac Studio, which offers up to 512GB of memory.
How much RAM do you need to run an LLM locally on Mac mini?
It depends on the model you need. The best model each Mac mini memory tier can run: 16GB: gpt-oss-20b (no AA v5 score yet); 24–32GB: Qwen3.8-27B (4-bit, tight on memory) (AA score 20.1); 48GB: Qwen3.8-27B (AA score 20.1); 64GB: Qwen3.8-27B (8-bit) (AA score 20.1). AA score: Artificial Analysis Intelligence Index v5.0, checked 3 Nov 2026 — higher means smarter; the v5 scale is harder, so scores are lower than before. For comparison, Claude Opus 5.5 scores 58.4 and the leader, Gemini 4 Carbon, scores 59.4. Response speed and the number of simultaneous users depend on the GPU and memory bandwidth; Namtech tests on real hardware before advising.
What is the best AI model Mac mini can run?
The 64GB version runs Qwen3.8-27B (8-bit) · AA score 20.1; Qwen3.8-27B at 8-bit takes about 29.5 GB, Apache 2.0 licence. The 16GB version runs gpt-oss-20b · no AA v5 score yet; gpt-oss-20b takes about 11 GB. For a smarter model, a 256GB Mac Studio runs GLM-5.3-Flash · AA score 27.8. AA score: Artificial Analysis Intelligence Index v5.0, checked 3 Nov 2026 — higher means smarter; the v5 scale is harder, so scores are lower than before. For comparison, Claude Opus 5.5 scores 58.4 and the leader, Gemini 4 Carbon, scores 59.4.
How much do the Mac mini M6 and M5 Pro cost?
Mac mini M6 is priced from 27,780,000₫ to 40,010,000₫, and M5 Pro from 53,340,000₫ to 90,000,000₫ with the default storage. Prices include VAT; the machines are sold and VAT-invoiced by Công ty Cổ phần Trạm Năng Lượng Số, and Namtech advises on choosing a configuration for the AI models you need to run. SSD upgrades and 10Gb Ethernet cost extra; choose them on each machine’s detail page.
What is the difference between Mac mini M6 and M5 Pro?
M5 Pro has 307GB/s of memory bandwidth versus 153GB/s – 170GB/s on M6, a 16- or 20-core GPU versus 12-core, and 24–64GB of memory (up to Qwen3.8-27B (8-bit) · AA score 20.1) versus 16–32GB (up to Qwen3.8-27B (4-bit, tight on memory) · AA score 20.1). On the back, M5 Pro has 3× Thunderbolt 5 (USB-C, up to 120Gb/s), while M6 has 3× Thunderbolt 4 (USB-C, up to 40Gb/s). M5 Pro storage goes up to 8TB, M6 up to 2TB.
Mac mini or Mac Studio: which should I choose for local AI?
Choose Mac mini for testing or personal use: 16–64GB memory (up to Qwen3.8-27B (8-bit) · AA score 20.1), from 27,780,000₫. Mac mini suits testing or personal use, not as the AI server of a Namtech internal-AI package. Choose Mac Studio when you need an AI server for a team, a smarter model or many people querying at once: 36–512GB memory (up to GLM-5.3-Flash · AA score 27.8), from 77,780,000₫. Prices include VAT.
Do I need the 10Gb Ethernet upgrade?
It is not required. Every Mac mini comes with a 2.5Gb Ethernet port; the 10Gb Ethernet option costs an extra 3,070,000₫ (VAT included) and must be selected when you order the machine. It is worth considering if the machine often copies large files over your internal network; for everyday text Q&A, the built-in port is usually enough.
Can I upgrade the RAM after purchase?
No. Mac mini memory is built into the chip, so you must choose the right amount when you order. That is why you should work out the models you plan to run before settling on a configuration.
What is the maximum storage for Mac mini?
M6 can be configured from 256GB to 2TB; M5 Pro from 512GB to 8TB. Storage holds the operating system, model files and documents; if you plan to keep many models or a large document library, choose extra capacity when you order.
Can Mac mini run an internal chatbot?
For testing or personal use, yes: Mac mini can run a document Q&A assistant with no data leaving. Mac mini suits testing or personal use, not as the AI server of a Namtech internal-AI package. A chatbot for the whole company should run on Mac Studio. Namtech can install and tune the software that runs your models (MLX, Ollama or LM Studio) to your requirements, quoted separately based on the scope of work.
Is a VAT invoice issued?
Yes. Prices listed on this page include VAT. Machines and accessories are sold by Công ty Cổ phần Trạm Năng Lượng Số, which issues full VAT invoices to businesses; Namtech advises on the configuration and installs the AI software. Configurations with upgraded RAM, SSD or Ethernet are built to order; we confirm the delivery date when we receive your request.