diff --git a/benchmarks/single_node/dsv4_fp4_b300_sglang_mtp.sh b/benchmarks/single_node/dsv4_fp4_b300_sglang_mtp.sh
index 767b9a8f9..d01f80a1d 100755
--- a/benchmarks/single_node/dsv4_fp4_b300_sglang_mtp.sh
+++ b/benchmarks/single_node/dsv4_fp4_b300_sglang_mtp.sh
@@ -128,6 +128,12 @@ wait_for_server_ready --port "$PORT" --server-log "$SERVER_LOG" --server-pid "$S
 
 pip install -q datasets pandas
 
+# --dsv4 routes prompts through encoding_dsv4.py (PR #1153), which emits the
+# <bos><User>...<Assistant><think> framing DeepSeek-V4-Pro expects. The DSv4-Pro
+# tokenizer ships without a jinja chat_template, so plain --use-chat-template
+# would crash; --dsv4 sidesteps that and satisfies the AGENTS.md rule that all
+# MTP scripts must benchmark against chat-formatted inputs (EAGLE acceptance
+# silently regresses on raw random tokens).
 run_benchmark_serving \
     --model "$MODEL" \
     --port "$PORT" \
@@ -138,7 +144,8 @@ run_benchmark_serving \
     --num-prompts $((CONC * 10)) \
     --max-concurrency "$CONC" \
     --result-filename "$RESULT_FILENAME" \
-    --result-dir "$PWD/"
+    --result-dir "$PWD/" \
+    --dsv4
 
 if [ "${RUN_EVAL}" = "true" ]; then
     run_eval --framework lm-eval --port "$PORT"
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 992c64ecb..2bd2f025c 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -1886,3 +1886,9 @@
     - "Image pinned to lmsysorg/sglang:deepseek-v4-blackwell@sha256:df18bfc4aa9ecf59451002b49ba00cae58042de9e2a96378bbd21b404dd62c7b"
     - "Adds SGLANG_OPT_* env knobs (SWA_SPLIT_LEAF_ON_INSERT, USE_JIT_NORM, USE_JIT_INDEXER_METADATA, USE_TOPK_V2, USE_CUSTOM_ALL_REDUCE_V2)"
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1187
+
+- config-keys:
+    - dsv4-fp4-b300-sglang-mtp
+  description:
+    - "Pass --dsv4 to run_benchmark_serving so MTP benchmarks use the DSv4 chat template (PR #1153)"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1182