# Qwen3.6-35B-A3B Q4_K_M on Ryzen AI Max+ 395 (llama.cpp ROCm) — inter-token latency: 18.7 ms/token

On a Beelink GTR9 Pro — AMD Ryzen AI Max+ 395, Radeon 8060S (gfx1151), 128 GB LPDDR5X-8000 unified running llama.cpp (server, ROCm backend), b9049 (server_fingerprint b9049-2496f9c14) with ggml-org/Qwen3.6-35B-A3B-GGUF @ baec3ebee244 (Q4_K_M), the measured inter-token latency was 18.7 ms/token (median over valid requests). 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

> Median inter-token latency (client-side decode-speed proxy) for Q4_K_M on the ROCm backend.

## Evidence

| Field | Value |
| --- | --- |
| Claim id | `strix.qwen36q4.rocm.c1.itl-nothink` |
| Value | 18.7 ms/token |
| Metric | inter-token latency |
| Aggregation | median over valid requests |
| System | Beelink GTR9 Pro — AMD Ryzen AI Max+ 395, Radeon 8060S (gfx1151), 128 GB LPDDR5X-8000 unified |
| Runtime | llama.cpp (server, ROCm 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 | 2026-08-03 |
| Last changed | 2026-08-03 |

**Limitations.** Three repeated runs on one unit, reasoning disabled.

## Raw evidence

- Sealed run bundle `run-20260803-rocm-v2-nothink-a`: https://github.com/botAGI/agmind-lab/tree/main/runs/run-20260803-rocm-v2-nothink-a
- Sealed run bundle `run-20260803-rocm-v2-nothink-a-r2`: https://github.com/botAGI/agmind-lab/tree/main/runs/run-20260803-rocm-v2-nothink-a-r2
- Sealed run bundle `run-20260803-rocm-v2-nothink-a-r3`: https://github.com/botAGI/agmind-lab/tree/main/runs/run-20260803-rocm-v2-nothink-a-r3
- Derivation query: https://github.com/botAGI/agmind-lab/blob/main/catalog/claims/sql/itl-median.sql
- Machine-readable: https://agmind.ai/claims/strix.qwen36q4.rocm.c1.itl-nothink.json · full registry https://agmind.ai/claims.json

## Cited on

- https://agmind.ai/reports/how-many-tokens-per-second-is-enough/
- https://agmind.ai/reports/q8-0-vs-q4-k-m-strix-halo/
- https://agmind.ai/reports/model-quant-backend-strix-halo/
- https://agmind.ai/compare/llamacpp-vulkan-vs-rocm-strix-halo/
- 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 ROCm): inter-token latency — 18.7 ms/token (median over valid requests; 3 runs on 1 unit; evidence level lab_repeated; workload interactive-assistant-v2@2026-08-02). Claim strix.qwen36q4.rocm.c1.itl-nothink. https://agmind.ai/claims/strix.qwen36q4.rocm.c1.itl-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.qwen36q4.rocm.c1.itl-nothink/