# Qwen3.6-35B-A3B Q4_K_M on Ryzen AI Max+ 395 (llama.cpp Vulkan) — time to first answer token, 95th percentile, 4 concurrent requests: 608 ms

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 time to first answer token, 95th percentile, 4 concurrent requests was 608 ms (95th percentile over valid requests, interpolated). The measurement was taken under the frozen interactive-assistant-v2@2026-08-02 workload; evidence level lab_repeated, 3 valid runs across 1 physical unit. The value is re-derived from the raw run records on every CI build.

## Statement

> 95th-percentile client-side time to the first answer token with the reasoning block disabled, while 4 closed-loop streams run concurrently on the unit (each request measured from its own client) — the slow tail of the same requests behind the median claim.

## Evidence

| Field | Value |
| --- | --- |
| Claim id | `strix.qwen36.interactive2.c4.ttfa-p95-nothink` |
| Value | 608 ms |
| Metric | time to first answer token, 95th percentile |
| Aggregation | 95th percentile over valid requests, interpolated |
| 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) |
| Workload scope | interactive-assistant-v2@2026-08-02 |
| Evidence level | lab_repeated |
| Status | active |
| Runs | 3 across 1 unit(s) |
| Published | 2026-09-02 |
| Last changed | 2026-09-02 |

**Limitations.** Three repeated runs on one unit. Closed-loop concurrency 4 — not an arrival-rate/SLO capacity claim. Context split across server slots (8192 per slot). Each run has 16 corpus items, so even pooled across the listed runs the 95th percentile rests on a handful of the slowest requests — a coarse tail statistic, not a tail-latency model. Failed requests carry no answer-token time and are outside this statistic; the completion-share claim on the same runs accounts for them.

## Raw evidence

- Sealed run bundle `run-20260803-v2-c4-nothink-a`: https://github.com/botAGI/agmind-lab/tree/main/runs/run-20260803-v2-c4-nothink-a
- Sealed run bundle `run-20260803-v2-c4-nothink-a-r2`: https://github.com/botAGI/agmind-lab/tree/main/runs/run-20260803-v2-c4-nothink-a-r2
- Sealed run bundle `run-20260803-v2-c4-nothink-a-r3`: https://github.com/botAGI/agmind-lab/tree/main/runs/run-20260803-v2-c4-nothink-a-r3
- Derivation query: https://github.com/botAGI/agmind-lab/blob/main/catalog/claims/sql/ttfa-p95-nothink.sql
- Machine-readable: https://agmind.ai/claims/strix.qwen36.interactive2.c4.ttfa-p95-nothink.json · full registry https://agmind.ai/claims.json

## Cited on

- https://agmind.ai/measured/qwen3-6-35b-a3b-q4-k-m-on-ryzen-ai-max-395/

## Cite this

AGmind Systems Lab. Qwen3.6-35B-A3B Q4_K_M on Ryzen AI Max+ 395 (llama.cpp Vulkan): time to first answer token, 95th percentile, 4 concurrent requests — 608 ms (95th percentile over valid requests, interpolated; 3 runs on 1 unit; evidence level lab_repeated; workload interactive-assistant-v2@2026-08-02). Claim strix.qwen36.interactive2.c4.ttfa-p95-nothink. https://agmind.ai/claims/strix.qwen36.interactive2.c4.ttfa-p95-nothink/

Registry changes feed: https://agmind.ai/claims/changes.xml · Corrections to published results: https://agmind.ai/errata/ · Data licensed CC BY 4.0.
Source page: https://agmind.ai/claims/strix.qwen36.interactive2.c4.ttfa-p95-nothink/