What should you buy to run AI models at home?
Live prices and stock for DGX Spark boxes, Strix Halo mini PCs, Mac Studios and big GPUs — matched to the open models each one can actually run.
| Machine | Lowest price | Stock | Memory | Bandwidth | $ / GB | Biggest models it runs well* |
|---|---|---|---|---|---|---|
| NVIDIA DGX Spark (Founders Edition, 4TB) DGX Spark / GB10 · CUDA |
$6,950 NVIDIA Marketplace |
Available | 128 GB LPDDR5x unified |
273 GB/s | $54.30 | Qwen3.5 122B-A10B Q4_K_M, Nemotron 3 Super 120B-A12B Q4_K_M, GPT-OSS 120B MXFP4 (native) |
| ASUS Ascent GX10 (1TB) DGX Spark / GB10 · CUDA |
$5,999 B&H Photo |
Out of stock | 128 GB LPDDR5x unified |
273 GB/s | $46.87 | Qwen3.5 122B-A10B Q4_K_M, Nemotron 3 Super 120B-A12B Q4_K_M, GPT-OSS 120B MXFP4 (native) |
| Gigabyte AI TOP ATOM (4TB) DGX Spark / GB10 · CUDA |
$5,300 B&H Photo |
Out of stock | 128 GB LPDDR5x unified |
273 GB/s | $41.41 | Qwen3.5 122B-A10B Q4_K_M, Nemotron 3 Super 120B-A12B Q4_K_M, GPT-OSS 120B MXFP4 (native) |
| Lenovo ThinkStation PGX (1TB) DGX Spark / GB10 · CUDA |
$5,859 B&H Photo |
Out of stock | 128 GB LPDDR5x unified |
273 GB/s | $45.77 | Qwen3.5 122B-A10B Q4_K_M, Nemotron 3 Super 120B-A12B Q4_K_M, GPT-OSS 120B MXFP4 (native) |
| HP ZGX Nano G1n (4TB) DGX Spark / GB10 · CUDA |
$7,699 B&H Photo |
Out of stock | 128 GB LPDDR5x unified |
273 GB/s | $60.15 | Qwen3.5 122B-A10B Q4_K_M, Nemotron 3 Super 120B-A12B Q4_K_M, GPT-OSS 120B MXFP4 (native) |
| GMKtec EVO-X2 (Ryzen AI Max+ 395, 128GB / 1TB) Strix Halo · ROCm / Vulkan |
$3,500 GMKtec store |
Available | 128 GB LPDDR5X-8000 unified |
256 GB/s | $27.34 | Qwen3.5 122B-A10B Q4_K_M, Nemotron 3 Super 120B-A12B Q4_K_M, GPT-OSS 120B MXFP4 (native) |
| GMKtec EVO-X2 (Ryzen AI Max+ 395, 64GB / 1TB) Strix Halo · ROCm / Vulkan |
$2,200 GMKtec store |
Available | 64 GB LPDDR5X-8000 unified |
256 GB/s | $34.37 | Qwen3.6 35B-A3B Q8_0, Gemma 4 31B Q4_K_M, GLM-4.7-Flash 30B-A3B Q8_0 |
| Framework Desktop DIY (Ryzen AI Max+ 395, 128GB) Strix Halo · ROCm / Vulkan |
$4,839 Framework store |
Out of stock | 128 GB LPDDR5x unified |
256 GB/s | $37.80 | Qwen3.5 122B-A10B Q4_K_M, Nemotron 3 Super 120B-A12B Q4_K_M, GPT-OSS 120B MXFP4 (native) |
| Minisforum MS-S1 Max (Ryzen AI Max+ 395, 128GB / 2TB) Strix Halo · ROCm / Vulkan |
$3,879 Minisforum store |
Available | 128 GB LPDDR5X-8000 unified |
256 GB/s | $30.30 | Qwen3.5 122B-A10B Q4_K_M, Nemotron 3 Super 120B-A12B Q4_K_M, GPT-OSS 120B MXFP4 (native) |
| Beelink GTR9 Pro (Ryzen AI Max+ 395, 128GB / 2TB) Strix Halo · ROCm / Vulkan |
$4,349 Beelink store |
Pre-order / delay | 128 GB LPDDR5X-8000 unified |
256 GB/s | $33.98 | Qwen3.5 122B-A10B Q4_K_M, Nemotron 3 Super 120B-A12B Q4_K_M, GPT-OSS 120B MXFP4 (native) |
| Mac Studio M5 Max (18-core CPU / 32-core GPU, 36GB) Mac Studio · MLX / Metal |
$2,499 Apple Store |
Available | 36 GB Unified memory |
460 GB/s | $69.42 | Qwen3.6 35B-A3B Q4_K_M, Gemma 4 31B Q4_K_M, GLM-4.7-Flash 30B-A3B Q4_K_M |
| Mac Studio M5 Max (18-core CPU / 40-core GPU, 48GB) Mac Studio · MLX / Metal |
$3,099 Apple Store |
Available | 48 GB Unified memory |
614 GB/s | $64.56 | Qwen3.6 35B-A3B Q4_K_M, Gemma 4 31B Q4_K_M, GLM-4.7-Flash 30B-A3B Q8_0 |
| Mac Studio M5 Ultra (30-core CPU / 64-core GPU, 96GB) Mac Studio · MLX / Metal |
$5,499 Apple Store |
Available | 96 GB Unified memory |
1,200 GB/s | $57.28 | GPT-OSS 120B MXFP4 (native), Llama 4 Scout 109B-A17B Q4_K_M, Qwen3-Next 80B-A3B Q4_K_M |
| Mac Studio M5 Ultra (256GB) Mac Studio · MLX / Metal |
unknown | Unknown | 256 GB Unified memory |
1,200 GB/s | — | Qwen3 235B-A22B Q4_K_M, Qwen3.5 122B-A10B Q8_0, Nemotron 3 Super 120B-A12B Q8_0 |
| Mac mini M5 Pro (15-core CPU / 16-core GPU, 24GB) Mac mini · MLX / Metal |
$1,699 Apple Store |
Available | 24 GB Unified memory |
307 GB/s | $70.79 | GPT-OSS 20B MXFP4 (native), Gemma 4 12B Q8_0, Qwen3.5 9B Q8_0 |
| NVIDIA GeForce RTX 5090 Founders Edition (32GB card) RTX 5090 · CUDA |
$8,500 Newegg |
Available | 32 GB GDDR7 VRAM |
1,792 GB/s | $265.62 | Qwen3.6 35B-A3B Q4_K_M, Gemma 4 31B Q4_K_M, GLM-4.7-Flash 30B-A3B Q4_K_M |
| Gigabyte GeForce RTX 4090 Windforce V2 (24GB card) RTX 4090 · CUDA |
$4,499 Newegg |
Available | 24 GB GDDR6X VRAM |
1,008 GB/s | $187.46 | Gemma 4 31B Q4_K_M, GLM-4.7-Flash 30B-A3B Q4_K_M, Qwen3.8 27B Q4_K_M |
| PNY NVIDIA RTX PRO 6000 Blackwell (96GB card) RTX PRO 6000 · CUDA |
$17,899 B&H Photo |
Available | 96 GB GDDR7 ECC VRAM |
1,792 GB/s | $186.45 | Qwen3.5 122B-A10B Q4_K_M, Nemotron 3 Super 120B-A12B Q4_K_M, GPT-OSS 120B MXFP4 (native) |
* Largest models that fit with an estimated ≥8 tokens/sec. GPU cards are card-only prices. How we estimate →
Popular comparisons
NVIDIA DGX Spark vs GMKtec EVO-X2NVIDIA DGX Spark vs Mac Studio M5 UltraASUS Ascent GX10 vs NVIDIA DGX SparkFramework Desktop DIY vs GMKtec EVO-X2NVIDIA GeForce RTX 5090 Founders Edition vs Mac Studio M5 MaxMinisforum MS-S1 Max vs Beelink GTR9 Pro
What hardware runs…
Qwen3.5 9BGemma 4 12BGPT-OSS 20BGemma 4 26B-A4BQwen3.8 27BGemma 4 31BGLM-4.7-Flash 30B-A3BQwen3.6 35B-A3BLlama 3.3 70BQwen3-Next 80B-A3BLlama 4 Scout 109B-A17BGPT-OSS 120BNemotron 3 Super 120B-A12BQwen3.5 122B-A10BQwen3 235B-A22BMuse Spark Open 816B-A42B
Get Best Local AI Rig restock & price-drop alerts
One weekly email: what dropped, what's back in stock, and which new open models your box can run. No spam.