Minisforum MS-S1 MAX 128GB / 4TB

What the numbers mean

By our fit rule about 108.8GB of the 128GB is usable for a model (85% of 128GB). A model fits when its 4-bit weights × 1.15 plus 2GB for context and runtime stay under that. That covers 14 of the 16 open models we list, up to gpt-oss-120b (about 63.4GB of weights at MXFP4). At 8-bit, 13 fit. Too big at 4-bit: Qwen3.8-Flash-Next (about 109.1GB) and GLM-5.3-Flash (about 194.8GB).

Measured on this chip (public results, not our own tests): Gemma 4 12B 29.3 tok/s (QAT UD-Q4_K_XL, Strix Halo Guide), Gemma 4 26B-A4B 54.2 tok/s (UD-Q4_K_M, Strix Halo Guide), Qwen3.8 27B 12.9 tok/s (Q4_K_M, Strix Halo Guide), Qwen3-Coder 30B-A3B 94.6 tok/s (UD-Q4_K_XL, Strix Halo Guide), Qwen3.6 35B-A3B 76.9 tok/s (Q4_K_M, Strix Halo Guide) and gpt-oss-120b 55.6 tok/s (MXFP4, Strix Halo Guide).

The US price has been $4,199 at all 8 daily checks since Oct 29, 2026. Since Oct 29, 2026 (8 days of checks) the EU price has ranged from €4,479 to €4,599; the low was last seen on Nov 2, 2026. The current €4,599 is the highest we have recorded.

Other machines we track with the same chip (AMD Ryzen AI Max+ 395 (Radeon 8060S)): GMKtec EVO-X2 128GB / 1TB ($3,500 / €3,350), GMKtec EVO-X2 128GB / 2TB ($3,650 / €3,500), GMKtec EVO-X2 64GB / 1TB ($2,200 / €2,000), GMKtec EVO-X3 128GB / 2TB ($3,800 / €3,600), Minisforum MS-S1 MAX 128GB / 2TB ($3,919 / €4,319), Minisforum MS-S1 MAX 64GB / 2TB ($2,599 / €2,799), Beelink GTR9 Pro 128GB / 2TB ($4,349 / ≈€3,875), Geekom A9 Mega 128GB / 2TB ($3,999 / €3,999), Framework Desktop 128GB ($3,449 / €3,889), Framework Desktop 64GB (≈$2,479 / €2,209) and HP Z2 Mini G1a (Ryzen AI Max+ PRO 395) (no tracked price). We use one set of public results per chip, so they show the same speeds here; the differences we can measure are price and stock. The cheapest 128GB machine we track in USD is Framework Desktop 128GB (Strix Halo) at $3,449, $750 less. On gpt-oss-120b speed per $1,000 it is number 10 of 14 machines with a price and a result (13.2 tok/s per $1k; see the ranking).

Before you buy

• US: read from the store's own product data at each check. In stock at the last check.
• EU: read from the store's own product data at each check. In stock at the last check.
• The gpt-oss-120b result was run on Ryzen AI Max+ 395, 128GB (Beelink GTR9 Pro). We apply it to every machine with the same chip.
• The Qwen3-Coder-30B-A3B result was run on AMD Ryzen AI Max+ 395 (Beelink GTR9 Pro). We apply it to every machine with the same chip.

This machine is listed on Cheapest 128GB machine for local AI, Best computer for running gpt-oss-120b and Best mini PC for local LLMs.