Measured
Qwen3.6-35B-A3B Q4_K_M on Ryzen AI Max+ 395 — every measured number
Everything this lab has measured for Qwen3.6-35B-A3B Q4_K_M on Ryzen AI Max+ 395: 32 claims across 6 frozen workloads, backends llama.cpp ROCm and llama.cpp Vulkan. Each number links to its permanent claim page with scope, limitations and sealed run evidence — values are re-derived from raw records in CI.
doc-session-v1@2026-08-03
| Metric | Value | Evidence | Claim |
|---|---|---|---|
| time to first token, first question over a 32k document · llama.cpp Vulkan | 33940 ms | lab_repeated | strix.qwen36.docsession.c1.ttft-q1-32k |
| time to first token, second question with prompt cache on · llama.cpp Vulkan | 860 ms | lab_repeated | strix.qwen36.docsession.c1.ttft-q2-32k-cache |
| time to first token, second question with prompt cache off · llama.cpp Vulkan | 33728 ms | lab_repeated | strix.qwen36.docsession.c1.ttft-q2-32k-nocache |
| time to first token, second question over an 8k document, cache on · llama.cpp Vulkan | 660 ms | lab_repeated | strix.qwen36.docsession.c1.ttft-q2-8k-cache |
endurance-30m-v1@2026-08-03
| Metric | Value | Evidence | Claim |
|---|---|---|---|
| request completion over three hours (concurrency 4) · llama.cpp Vulkan | 100.0 % of requests | lab_single_run | strix.qwen36.endurance.c4.completion-180m |
| decode-pace drift over three hours (concurrency 4) · llama.cpp Vulkan | 1.6 % | lab_single_run | strix.qwen36.endurance.c4.itl-drift-180m |
| inter-token latency (concurrency 4) · llama.cpp Vulkan | 30.0 ms/token | lab_single_run | strix.qwen36.endurance.c4.itl-median |
interactive-assistant-v1@2026-08-02
| Metric | Value | Evidence | Claim |
|---|---|---|---|
| answerless (empty) responses at a 1k budget · llama.cpp Vulkan | 75.0 % of requests | lab_repeated | strix.qwen36.interactive.c1.answerless-1k |
| time to first answer token · llama.cpp Vulkan | 220 ms | lab_repeated | strix.qwen36.interactive.c1.ttfa-nothink |
| time to first answer token with reasoning on · llama.cpp Vulkan | 23453 ms | lab_repeated | strix.qwen36.interactive.c1.ttfa-thinking |
| time to the first token of any output · llama.cpp Vulkan | 221 ms | lab_repeated | strix.qwen36.interactive.c1.ttft-any-token |
interactive-assistant-v2@2026-08-02
| Metric | Value | Evidence | Claim |
|---|---|---|---|
| answerless (empty) responses at a 1k budget · llama.cpp Vulkan | 58.3 % of requests | lab_unit_replicated | strix.qwen36.interactive2.c1.answerless-1k |
| request completion · llama.cpp Vulkan | 100.0 % of requests | lab_unit_replicated | strix.qwen36.interactive2.c1.completion-nothink |
| time to first answer token · llama.cpp Vulkan | 210 ms | lab_unit_replicated | strix.qwen36.interactive2.c1.ttfa-nothink |
| time to first answer token with reasoning on · llama.cpp Vulkan | 20191 ms | lab_unit_replicated | strix.qwen36.interactive2.c1.ttfa-thinking |
| time to the first token of any output · llama.cpp Vulkan | 212 ms | lab_unit_replicated | strix.qwen36.interactive2.c1.ttft-any-token |
| request completion (concurrency 4) · llama.cpp Vulkan | 100.0 % of requests | lab_repeated | strix.qwen36.interactive2.c4.completion-nothink |
| time to first answer token (concurrency 4) · llama.cpp Vulkan | 338 ms | lab_repeated | strix.qwen36.interactive2.c4.ttfa-nothink |
| request completion (concurrency 8) · llama.cpp Vulkan | 100.0 % of requests | lab_repeated | strix.qwen36.interactive2.c8.completion-nothink |
| time to first answer token (concurrency 8) · llama.cpp Vulkan | 865 ms | lab_repeated | strix.qwen36.interactive2.c8.ttfa-nothink |
| inter-token latency · llama.cpp ROCm | 18.7 ms/token | lab_repeated | strix.qwen36q4.rocm.c1.itl-nothink |
| time to first answer token · llama.cpp ROCm | 215 ms | lab_repeated | strix.qwen36q4.rocm.c1.ttfa-nothink |
| inter-token latency · llama.cpp Vulkan | 15.9 ms/token | lab_repeated | strix.qwen36q4.vulkan.c1.itl-nothink |
long-context-v1@2026-08-03
| Metric | Value | Evidence | Claim |
|---|---|---|---|
| unanswerable-control honesty · llama.cpp Vulkan | 100.0 % of requests | lab_unit_replicated | strix.qwen36.longctx.c1.control-success |
| needle retrieval success · llama.cpp Vulkan | 100.0 % of requests | lab_repeated | strix.qwen36.longctx.c1.needle-success |
| time to first token at a 2k-token document · llama.cpp Vulkan | 1907 ms | lab_repeated | strix.qwen36.longctx.c1.ttft-2k-en |
| time to first token at a 32k-token document · llama.cpp Vulkan | 33965 ms | lab_repeated | strix.qwen36.longctx.c1.ttft-32k-en |
structured-agent-v1@2026-08-03
| Metric | Value | Evidence | Claim |
|---|---|---|---|
| end-to-end time to a complete answer · llama.cpp Vulkan | 844 ms | lab_repeated | strix.qwen36.structured.c1.e2e-nothink |
| end-to-end time with reasoning on · llama.cpp Vulkan | 14677 ms | lab_repeated | strix.qwen36.structured.c1.e2e-think4k |
| strict-JSON task success · llama.cpp Vulkan | 100.0 % of requests | lab_repeated | strix.qwen36.structured.c1.task-success-nothink |
| strict-JSON task success with reasoning on · llama.cpp Vulkan | 100.0 % of requests | lab_repeated | strix.qwen36.structured.c1.task-success-think4k |
| strict-JSON task success · llama.cpp ROCm | 100.0 % of requests | lab_repeated | strix.qwen36q4.rocm.c1.task-success |
How to read this page
These are scoped measurements, not a ranking. A verdict for a workload lives on its tested-configuration card; a number without its limitations is not comparable across systems. Statements are EN-canonical on the claim pages.