AMD Ryzen AI Max+ 395 in a Mac Studio–shaped 180 mm cube. 128GB soldered LPDDR5X, dual 10GbE, dual USB4, vapor-chamber cooling. Same chip class as Framework and EVO-X2, with networking as the differentiator.
Street price (USD)
$3,600–$4,350
Memory
128GB LPDDR5X soldered
Bandwidth
~256 GB/s
Chip
Ryzen AI Max+ 395
Catch
Soldered; price rose; check NIC revision
Clip
Beelink GTR9 Pro — Ryzen AI Max+ 395 with up to 128GB and the Radeon 8060S.
ETA PRIME, 2025-09-21. 00:02–00:24.
What it is
Beelink’s Strix Halo mini. 16 Zen 5 cores, Radeon 8060S (40 RDNA 3.5 CUs), 128GB of soldered LPDDR5X-8000 on a ~256 GB/s bus, dual M.2 PCIe 4.0, dual 10GbE, dual USB4 (40 Gbps), Wi-Fi 7, internal 230W PSU. Cooling is dual turbine fans plus vapor chamber, rated for 140W at claimed ~32 dB. Ships with 2TB SSD; second slot free. Only the 128GB configuration is the relevant SKU for local models.
Who it is for
People who want a 24B–70B local model plus dual 10GbE in one sealed cube, and who are fine with ROCm / llama.cpp / Vulkan instead of CUDA. Useful if media or models live on a NAS. Wrong if you need repairability, full discrete-GPU bandwidth, or the lowest possible price for this silicon.
What it actually runs
Comfortable: 24B–32B dense at usable speed; smaller MoE models at high token rates.
Possible: 70B-class dense and larger sparse MoE (e.g. 120B MoE reported in the 30–55 tok/s range on tuned Linux setups). Capacity is the win; decode is still limited by the ~256 GB/s bus. Dense 70B lands nearer single-digit tokens per second.
Not the win: training, CUDA-only stacks, or expecting discrete-GPU token rates. Ignore vendor “126 TOPS” and DeepSeek-70B marketing slides.
Standout details
Price snapshot early Sep 2026: official direct listings around $4,349 (CAD-equivalent markets similar); Amazon and third-party listings tracked in the ~$3,600–$3,650 range when available. Launch window was ~$1,985–$2,000. Stock and list prices move.
Memory is soldered LPDDR5X-8000. Up to ~96GB allocatable to the iGPU on Windows; higher effective pool on Linux. Pick 128GB or do not buy this for large models.
Dual 10GbE is the real differentiator versus sealed Strix Halo minis that ship 2.5GbE. Board revisions matter: early v1.0 used Intel E610 (some sustained-load reports); later v2.2 uses Realtek. Confirm which is shipping.
Quiet under load in controlled tests; a few reports of higher fan noise depending on ambient and sustained power. Firmware and driver rough edges still appear in early-unit notes.
Honest verdict
The dual-10GbE Strix Halo cube. Buy it if you specifically need the networking and 128GB in a sealed form factor at current street prices. Same bandwidth ceiling as every other Max+ 395 box. If you want repairability, get Framework. If you want a PCIe slot, look at MS-S1 MAX. If you want CUDA, shop Spark / GX10.
Prices and stock move. Affiliate links stay off until this directory has real traffic.