No single device we track can hold it. Full answer →
What should I buy to run AI at home, priced in Australia today?
GPUs, Macs and AI mini PCs ranked by how much model they can hold and how fast they generate text per dollar. Every price was pulled from an Australian retailer page, with the link and the time it was checked. Every speed is either a cited public benchmark or clearly labelled as an estimate.
27 of 34 items with a live AUD price · 27 with a measured benchmark · refreshed 20 Oct 2026, 7:38 am AEST
Trending this week week of 19 Oct 2026
New and most-talked-about open models, and what you'd need to run each one. Sourced from the Hugging Face trending list, Ollama's library and posts on X (links on every card). Updated weekly.
DeepSeek V4.1 Flash 1017
DeepSeek's refreshed V4.1 Flash. A 763B-parameter download (incl. Engram n-gram tables). Community posts say mainline llama.cpp support isn't there yet and a Q2_K GGUF is ~264 GB, so this is a multi-hundred-GB-rig model.
763.21B (MoE, active not published) · Q4 needs ~464 GB · no single device we track fits
Hugging Face · DeepSeek launch post (X) · @localllmrig on GGUF sizes (X) · GGUF: unsloth
Meta Muse Spark Open 816B-A42B
Meta's open-weights frontier MoE (816B total / 42B active, 1M context, per the model card). Gated download. X posts say 'at home' means two maxed-out M5 Ultra Mac Studios.
822.04B (42B active, MoE) · Q4 needs ~499 GB+ · no single device we track fits
Hugging Face · X post: needs two M5 Ultras · GGUF: unsloth · GGUF: ggml-org
Hy Memory 30B A3B (MoE)
New from Tencent: 30.06B parameters, 3.0B active (mixture-of-experts). No GGUF build found yet.
30.06B (3B active, MoE) · Q4 needs ~20 GB · cheapest in-stock fit: AMD Radeon AI PRO R9700 32GB A$2,299
Ling 3.1 flash (MoE)
New from Ling: 572.12B parameters, MoE (active count not published). No GGUF build found yet.
572.12B (25B active, MoE) · Q4 needs ~349 GB · no single device we track fits
Liquid LFM2.5 24B-A2B
24B MoE with only 2B active, built for on-device agents. Small enough for a 16GB card at Q4, and fast because so few weights are read per token.
23.97B (2B active, MoE) · Q4 needs ~16 GB · cheapest in-stock fit: AMD Radeon AI PRO R9700 32GB A$2,299
MiniCPM V 4.7 35B A3B (MoE)
New from MiniCPM: 35.21B parameters, 3.0B active (mixture-of-experts). GGUF builds available (openbmb).
35.21B (3B active, MoE) · Q4 needs ~23 GB · cheapest in-stock fit: AMD Radeon AI PRO R9700 32GB A$2,299
Qwen3.8 27B
New from Qwen: 27.78B parameters. GGUF builds available (unsloth, ggml-org, lmstudio-community).
27.78B · Q4 needs ~18 GB · cheapest in-stock fit: AMD Radeon AI PRO R9700 32GB A$2,299
MiniCPM V 4.7 1B
New from MiniCPM: 1.55B parameters. GGUF builds available (openbmb).
1.55B · Q4 needs ~2 GB · cheapest in-stock fit: Intel Arc B580 12GB A$419
Hardware news
- RTX Spark laptops land in Australia: Surface Laptop Ultra (NVIDIA RTX Spark N1X, up to 128GB unified memory) is now on Microsoft Store AU. We track its 64GB and 128GB configs. Surface Laptop Ultra (RTX Spark N1X, 6144-core GPU) 128GB A$9,649 · Surface Laptop Ultra (RTX Spark N1X, 6144-core GPU) 64GB A$7,149 (Microsoft Store AU · NVIDIA specs)
- RTX 5090 supply scare: Reports said NVIDIA stopped allocating GB202 dies to the RTX 5090. Wccftech then reported that NVIDIA denied it. Expect volatile AU pricing. (Wccftech: NVIDIA denies discontinuation (X) · X story summary)
- RAM prices keep climbing: Posts cite consumer DDR5 kits up sharply since September and contract DRAM prices more than doubling. This matters because MoE models are often run with experts offloaded to system RAM. (X post on DDR5 pricing)
Can I run it?
Pick a model, quantisation and context length. We'll show which hardware can hold it in GPU memory and roughly how fast it will run.
| Hardware | Needs / usable | Fit | Est. tok/s | Price |
|---|
The memory needed is weights + KV cache + 1GB overhead. Speed is estimated from memory bandwidth. See how we calculate it.
Hardware table
Click a column heading to sort.Use the menu below to sort. Speed and value use the reference test Llama 2 7B Q4_0 (llama.cpp, tg128).
| Hardware | Memory | Usable for AI | Bandwidth | Price (AUD) | tok/s (ref) | tok/s per A$1k | Biggest model (Q4, 8k ctx) |
|---|---|---|---|---|---|---|---|
Intel Arc B580 12GB Graphics card |
12 GB VRAM |
11.5 GB |
456 GB/s |
A$419 Scorptec checked 20 Oct 2026, 7:44 am AEST |
70 measured src |
167.4 |
|
NVIDIA RTX 3060 12GB Graphics card |
12 GB VRAM |
11.5 GB |
360 GB/s |
A$539 Scorptec checked 20 Oct 2026, 7:38 am AEST |
76 measured src |
140.2 |
|
AMD Radeon RX 9070 XT 16GB Graphics card |
16 GB VRAM |
15.5 GB |
640 GB/s |
A$1,169 Mwave checked 20 Oct 2026, 7:38 am AEST |
137 measured src |
117.3 |
|
AMD Radeon RX 9070 16GB Graphics card |
16 GB VRAM |
15.5 GB |
640 GB/s |
A$1,099 Mwave checked 20 Oct 2026, 7:38 am AEST |
120 measured src |
108.9 |
|
NVIDIA RTX 5070 Ti 16GB Graphics card |
16 GB VRAM |
15.5 GB |
896 GB/s |
A$1,699 Scorptec checked 20 Oct 2026, 7:40 am AEST |
177 measured src |
104.1 |
|
NVIDIA RTX 5070 12GB Graphics card |
12 GB VRAM |
11.5 GB |
672 GB/s |
A$1,249 Scorptec checked 20 Oct 2026, 7:39 am AEST |
128 measured src |
102.1 |
|
AMD Radeon RX 9060 XT 16GB Graphics card |
16 GB VRAM |
15.5 GB |
320 GB/s |
A$699 Mwave checked 20 Oct 2026, 7:38 am AEST |
68 measured src |
96.7 |
|
NVIDIA RTX 5060 Ti 16GB Graphics card |
16 GB VRAM |
15.5 GB |
448 GB/s |
A$1,099 Scorptec checked 20 Oct 2026, 7:39 am AEST |
91 measured src |
82.7 |
|
NVIDIA RTX 5080 16GB Graphics card |
16 GB VRAM |
15.5 GB |
960 GB/s |
A$2,299 Scorptec checked 20 Oct 2026, 7:40 am AEST |
182 measured src |
79.2 |
|
AMD Radeon AI PRO R9700 32GB Graphics card |
32 GB VRAM |
31.5 GB |
640 GB/s |
A$2,299 Mwave checked 20 Oct 2026, 7:38 am AEST |
146 measured src |
63.4 |
|
NVIDIA RTX 3090 Ti 24GB Graphics card |
24 GB VRAM |
23.5 GB |
1,008 GB/s |
A$3,599 Scorptec · not in stock online checked 20 Oct 2026, 7:41 am AEST |
171 measured src |
47.6 |
|
NVIDIA RTX 5090 32GB Graphics card |
32 GB VRAM |
31.5 GB |
1,792 GB/s |
A$10,999 Scorptec checked 20 Oct 2026, 7:41 am AEST |
290 measured src |
26.4 |
|
24 GB unified |
18.0 GB |
307 GB/s |
A$2,699 Apple AU checked 20 Oct 2026, 7:44 am AEST |
68 measured src |
25.0 |
||
96 GB unified |
72.0 GB |
1,200 GB/s |
A$9,499 Apple AU checked 20 Oct 2026, 7:44 am AEST |
234 estimate |
24.7 |
||
24 GB unified |
18.0 GB |
307 GB/s |
A$2,999 Apple AU checked 20 Oct 2026, 7:44 am AEST |
66 measured src |
22.1 |
||
36 GB unified |
27.0 GB |
460 GB/s |
A$4,299 Apple AU checked 20 Oct 2026, 7:44 am AEST |
90 estimate |
20.9 |
||
16 GB unified |
12.0 GB |
153 GB/s |
A$1,449 Apple AU checked 20 Oct 2026, 7:44 am AEST |
30 estimate |
20.6 |
||
256 GB unified |
192.0 GB |
1,200 GB/s |
A$11,449 Apple AU checked 20 Oct 2026, 7:44 am AEST |
179 measured src |
15.6 |
||
16 GB unified |
12.0 GB |
120 GB/s |
A$1,597 Mwave checked 20 Oct 2026, 7:44 am AEST |
24 measured src |
15.1 |
||
64 GB unified |
48.0 GB |
256 GB/s |
A$3,915 Minisforum AU · not in stock online checked 20 Oct 2026, 7:44 am AEST |
50 measured src |
12.8 |
||
NVIDIA RTX PRO 6000 Blackwell 96GB Graphics card |
96 GB VRAM |
95.5 GB |
1,792 GB/s |
A$27,599 Scorptec checked 20 Oct 2026, 7:42 am AEST |
274 measured src |
9.9 |
|
128 GB unified |
96.0 GB |
256 GB/s |
A$5,719 Minisforum AU checked 20 Oct 2026, 7:44 am AEST |
50 measured src |
8.7 |
||
64 GB unified |
48.0 GB |
256 GB/s |
A$5,894 Mwave checked 20 Oct 2026, 7:44 am AEST |
50 estimate |
8.5 |
||
64 GB unified |
48.0 GB |
301 GB/s |
A$7,149 Microsoft Store AU · stale checked 19 Oct 2026, 8:18 pm AEST |
59 estimate |
8.2 |
||
NVIDIA DGX Spark (GB10) 128GB AI mini PC |
128 GB unified |
96.0 GB |
273 GB/s |
A$8,999 Mwave checked 20 Oct 2026, 7:44 am AEST |
57 measured src |
6.4 |
|
128 GB unified |
96.0 GB |
301 GB/s |
A$9,649 Microsoft Store AU · stale checked 19 Oct 2026, 8:18 pm AEST |
59 estimate |
6.1 |
||
128 GB unified |
96.0 GB |
256 GB/s |
A$8,999 Mwave · not in stock online checked 20 Oct 2026, 7:44 am AEST |
50 measured src |
5.6 |
||
NVIDIA RTX 4060 Ti 16GB Graphics card |
16 GB VRAM |
15.5 GB |
288 GB/s |
price unavailable |
56 estimate |
– |
|
NVIDIA RTX 3090 24GB Graphics card |
24 GB VRAM |
23.5 GB |
936 GB/s |
price unavailable |
158 measured src |
– |
|
NVIDIA RTX 4090 24GB Graphics card |
24 GB VRAM |
23.5 GB |
1,008 GB/s |
price unavailable |
186 measured src |
– |
|
AMD Radeon RX 7900 XTX 24GB Graphics card |
24 GB VRAM |
23.5 GB |
960 GB/s |
price unavailable |
183 measured src |
– |
|
48 GB unified |
36.0 GB |
614 GB/s |
price unavailable |
120 measured src |
– |
||
GMKtec EVO-X2 (Ryzen AI Max+ 395) 64GB AI mini PC |
64 GB unified |
48.0 GB |
256 GB/s |
price unavailable |
50 measured src |
– |
|
128 GB unified |
96.0 GB |
256 GB/s |
price unavailable |
50 measured src |
– |