Measured

Measured pairs: every model on every system the lab has numbers for

One page per model × system pairing with at least three registry claims. Each page lists every measured number for that pairing, grouped by frozen workload, and links to the claim pages with scope, limitations and sealed runs. Pairings not listed here have not been measured.

Qwen3.6-35B-A3B Q4_K_M on Ryzen AI Max+ 395
38 claims across 6 frozen workloads · llama.cpp ROCm, llama.cpp Vulkan
gemma-4-26B-A4B-it Q4_0 on Ryzen AI Max+ 395
6 claims across 3 frozen workloads · llama.cpp Vulkan

This is coverage, not a ranking. The list grows only when new claims enter the registry; values are re-derived from raw run records in CI.