diff --git a/benchmarks/single_node/agentic/qwen3.5_fp4_mi355x_sglang_mtp.sh b/benchmarks/single_node/agentic/qwen3.5_fp4_mi355x_sglang_mtp.sh
index e9e3d5eb5..9a09392e9 100644
--- a/benchmarks/single_node/agentic/qwen3.5_fp4_mi355x_sglang_mtp.sh
+++ b/benchmarks/single_node/agentic/qwen3.5_fp4_mi355x_sglang_mtp.sh
@@ -14,7 +14,7 @@ check_env_vars \
     MODEL TP CONC EP_SIZE KV_OFFLOADING \
     TOTAL_CPU_DRAM_GB RESULT_DIR DURATION
 
-SCHEDULER_RECV_INTERVAL=${SCHEDULER_RECV_INTERVAL:-30}
+SCHEDULER_RECV_INTERVAL=${SCHEDULER_RECV_INTERVAL:-60}
 
 if [[ -n "${SLURM_JOB_ID:-}" ]]; then
     echo "JOB $SLURM_JOB_ID running on ${SLURMD_NODENAME:-unknown}"
@@ -90,7 +90,7 @@ export SGLANG_USE_AITER=1
 export SGLANG_USE_AITER_UNIFIED_ATTN=1
 export AITER_FLYDSL_FORCE=1
 export SGLANG_MAMBA_SSM_DTYPE=bfloat16
-export ROCM_QUICK_REDUCE_QUANTIZATION=INT8
+export ROCM_QUICK_REDUCE_QUANTIZATION=INT4
 export SGLANG_TIMEOUT_KEEP_ALIVE=1800
 
 if [ "${EVAL_ONLY:-false}" != "true" ]; then
@@ -113,7 +113,7 @@ SGLANG_CMD=(
     --watchdog-timeout 1200
     --page-size 16
     --kv-cache-dtype fp8_e4m3
-    --cuda-graph-max-bs "$CUDA_GRAPH_MAX_BS"
+    --cuda-graph-max-bs-decode "$CUDA_GRAPH_MAX_BS"
     --max-running-requests "$MAX_RUNNING_REQUESTS"
     --max-prefill-tokens 16384
     --chunked-prefill-size 16384
diff --git a/configs/amd-master.yaml b/configs/amd-master.yaml
index 26010d293..4958de065 100644
--- a/configs/amd-master.yaml
+++ b/configs/amd-master.yaml
@@ -361,7 +361,7 @@ qwen3.5-fp4-mi355x-sglang-mtp:
       - { tp: 4, conc-start: 4, conc-end: 16, spec-decoding: mtp }
 
 qwen3.5-fp4-mi355x-sglang-agentic-mtp:
-  image: lmsysorg/sglang-rocm:v0.5.19-rocm720-mi35x-20260908
+  image: lmsysorg/sglang-rocm:v0.5.19-rocm720-mi35x-20260911
   model: amd/Qwen3.5-397B-A17B-MXFP4
   model-prefix: qwen3.5
   runner: cluster:mi355x-amds
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index c69451c2e..b65157b98 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -7561,3 +7561,13 @@
     - "Replace the local 2*CONC max_num_seqs limit with an explicit reproducibility cap of 128 (the effective vllm serve default on MI355X is 1024). The old cap sat below AgentX subagent fan-out and produced queue times up to 16.2 seconds at concurrency 1; 128 removed that scheduling queue. The derived CUDA-graph capture ceiling is 1024 for each configured concurrency."
     - "将本地 2*CONC 的 max_num_seqs 上限改为显式固定的可复现值 128（MI355X 上 vllm serve 的实际默认值为 1024）。旧上限低于 AgentX 子代理扇出，并发 1 时排队时间最高达 16.2 秒；128 消除了该调度排队。所有已配置并发点由此得到的 CUDA graph 捕获上限均为 1024。"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/3058
+
+- config-keys:
+    - qwen3.5-fp4-mi355x-sglang-agentic-mtp
+  scenario-type:
+    - agentic-coding
+  description:
+    - "Bump image to lmsysorg/sglang-rocm:v0.5.19-rocm720-mi35x-20260911."
+    - "Rename --cuda-graph-max-bs to --cuda-graph-max-bs-decode for sglang 0.5.19."
+    - "Prefill allreduce INT4; scheduler-recv-interval default 60."
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/2972