Qwen3.6-35B-A3B Q4_K_M on Ryzen AI Max+ 395 (llama.cpp Vulkan) — run-to-run spread of the median time to first answer token: 11 ms

The measurement

On a Beelink GTR9 Pro — AMD Ryzen AI Max+ 395, Radeon 8060S (gfx1151), 128 GB LPDDR5X-8000 unified running llama.cpp (server, Vulkan backend), b9049 (server_fingerprint b9049-2496f9c14) with ggml-org/Qwen3.6-35B-A3B-GGUF @ baec3ebee244 (Q4_K_M), the measured run-to-run spread of the median time to first answer token was 11 ms (highest minus lowest per-run median over valid requests, across the listed runs). The measurement was taken under the frozen interactive-assistant-v2@2026-08-02 workload; evidence level lab_unit_replicated, 6 valid runs across 2 physical units. The value is re-derived from the raw run records on every CI build.

Statement

Spread between the per-run medians of client-side time to the first answer token, reasoning block disabled, across the listed runs and units: the highest run median minus the lowest. A repeatability indicator for this cell, not a variance model.

Evidence

System

Beelink GTR9 Pro — AMD Ryzen AI Max+ 395, Radeon 8060S (gfx1151), 128 GB LPDDR5X-8000 unified

Runtime

llama.cpp (server, Vulkan backend), b9049 (server_fingerprint b9049-2496f9c14)

Model artifact

ggml-org/Qwen3.6-35B-A3B-GGUF @ baec3ebee244 (Q4_K_M)

Scope

interactive-assistant-v2@2026-08-02

Aggregation

highest minus lowest per-run median over valid requests, across the listed runs

Evidence level

lab_unit_replicated

Status

active

Published

September 2, 2026

Limitations

Three repeated runs across two commercially identical units — six per-run medians in total, so the spread is a range over a small sample, not a confidence interval. Pools unit-to-unit and repeat-to-repeat differences into one number; it does not separate the two. Each run has 16 corpus items, so a single per-run median moves with one or two requests.

Runs

• run-20260802-v2-nothink-a
• run-20260802-v2-nothink-a-r2
• run-20260802-v2-nothink-a-r3
• run-20260803-v2-nothink-b
• run-20260803-v2-nothink-b-r2
• run-20260803-v2-nothink-b-r3

Derivation

The value is re-derived from the raw run records by this query on every CI build — a published number cannot silently drift from its evidence.

catalog/claims/sql/ttfa-run-spread.sql

Cited on

Pages on this site that render this number:

Cite this claim

AGmind Systems Lab. Qwen3.6-35B-A3B Q4_K_M on Ryzen AI Max+ 395 (llama.cpp Vulkan): run-to-run spread of the median time to first answer token — 11 ms (highest minus lowest per-run median over valid requests, across the listed runs; 6 runs on 2 units; evidence level lab_unit_replicated; workload interactive-assistant-v2@2026-08-02). Claim strix.qwen36.interactive2.c1.ttfa-nothink-spread. https://agmind.ai/claims/strix.qwen36.interactive2.c1.ttfa-nothink-spread/

@misc{agmind_strix_qwen36_interactive2_c1_ttfa_nothink_spread,
  author       = {{AGmind Systems Lab}},
  title        = {Spread between the per-run medians of client-side time to the first answer token, reasoning block disabled, across the listed runs and units: the highest run median minus the lowest. A repeatability indicator for this cell, not a variance model.},
  howpublished = {\url{https://agmind.ai/claims/strix.qwen36.interactive2.c1.ttfa-nothink-spread/}},
  note         = {Claim strix.qwen36.interactive2.c1.ttfa-nothink-spread: 11 ms; evidence level lab\_unit\_replicated; scope interactive-assistant-v2@2026-08-02},
  year         = {2026}
}

Machine-readable: /claims/strix.qwen36.interactive2.c1.ttfa-nothink-spread.json · BibTeX · CSL-JSON · full registry · changes feed