# Framework Desktop (Ryzen AI Max+ 395) for local AI

URL: https://insidethemachine.ai/machines/framework-desktop-395
Type: AMD mini PC. Chip: AMD Ryzen AI Max+ 395 (Strix Halo, Radeon 8060S 40CU). Memory: unified, options 64, 128 GB. Bandwidth: 256 GB/s (256-bit LPDDR5X-8000 → 256 GB/s theoretical.). Spec source: https://openclawdc.com/blog/is-a-strix-halo-mini-pc-worth-it/ (2026-09-23).
Price: from about $1,959 (2026-10-15, approximate). Status: current.

## Which models fit

| memory GB | usable GB | models that fit | best 3 |
|---|---|---|---|
| 64 | 48 | 77 | Qwen3.8 27B (22.7 GB), K2 Horizon MoVA 36B-A4B (28.1 GB), Qwen3.5 27B (21.5 GB) |
| 128 | 96 | 90 | Qwen3.8 27B (22.7 GB), K2 Horizon MoVA 36B-A4B (28.1 GB), Qwen3.5 27B (21.5 GB) |

## Published speeds (not our tests)

- gpt-oss-120b: 54 tok/s decode, 500 tok/s prompt (llama.cpp Vulkan, MXFP4 (GGUF), Framework Desktop 128GB (llama-bench pp512/tg128), community-measured) https://llamaperf.com/model/gpt-oss (2026-10-15)

Notes: Launched at $1,999 (128GB); ~73% higher now due to DRAM prices. Only Strix Halo box with a PCIe x4 slot. GPU can use ~96GB of 128GB by default (sunkcost.ai).

## Questions

**What is the best local LLM for the Framework Desktop (Ryzen AI Max+ 395)?**
With 128 GB, Qwen3.8 27B (needs 22.7 GB, AA 33.7). 90 of 124 models fit. With 64 GB: Qwen3.8 27B.

**How fast is the Framework Desktop (Ryzen AI Max+ 395) for local AI?**
Published: gpt-oss-120b at 54 tokens/sec (llama.cpp Vulkan, Framework Desktop 128GB (llama-bench pp512/tg128)). These are other people's measurements, not ours.

**How much does the Framework Desktop (Ryzen AI Max+ 395) cost?**
From about $1,959 (as of 2026-10-15). 2026 prices are inflated by the memory-chip shortage; check before buying.