verifiedmedium · 0.75updated 2026-10-14

Ryzen AI Max+ 395 Strix Halo GLM-4.5-Air UD-Q4_K_XL decode

23.4 tok/s (single-stream)

Fields

benchmarks/comm-strix-halo-glm45-air-udq4-decode

Metric

decode_tok_s

Value

23.4

Unit

tok/s (single-stream)

Model

GLM-4.5-Air UD-Q4_K_XL (106B, ~12B active)

Methodology

lhl/strix-halo-testing llm-bench, Vulkan AMDVLK; ~68 GB resident

Measured by

third_party

Thermal state

unknown