diff --git a/.github/configs/nvidia-master.yaml b/.github/configs/nvidia-master.yaml
index 9d99372b2..05b19d802 100644
--- a/.github/configs/nvidia-master.yaml
+++ b/.github/configs/nvidia-master.yaml
@@ -1706,6 +1706,27 @@ dsv4-fp4-b200-sglang:
     # DP-attention (DP_ATTENTION=true) — max-throughput CONC range
     - { tp: 8, ep: 8, dp-attn: true, conc-start: 256, conc-end: 512 }
 
+dsv4-fp4-b200-vllm:
+  image: vllm/vllm-openai:deepseekv4-cu130
+  model: deepseek-ai/DeepSeek-V4-Pro
+  model-prefix: dsv4
+  runner: b200-dsv4
+  precision: fp4
+  framework: vllm
+  multinode: false
+  seq-len-configs:
+  - isl: 1024
+    osl: 1024
+    search-space:
+    - { tp: 8, conc-start: 1, conc-end: 64 }
+    - { tp: 8, ep: 8, conc-start: 128, conc-end: 128 }
+    - { tp: 8, ep: 8, dp-attn: true, conc-start: 256, conc-end: 4096 }
+  - isl: 8192
+    osl: 1024
+    search-space:
+    - { tp: 8, conc-start: 1, conc-end: 32 }
+    - { tp: 8, ep: 8, dp-attn: true, conc-start: 64, conc-end: 1024 }
+
 # NOTE: At the time of submission, https://cookbook.sglang.io/autoregressive/DeepSeek/DeepSeek-R1
 # does not have a B300-specific recipe, so this config reuses the existing DSR1 FP4
 # B200 SGLang recipe as-is until B300-specific tuning is available.
diff --git a/benchmarks/single_node/dsv4_fp4_b200_vllm.sh b/benchmarks/single_node/dsv4_fp4_b200_vllm.sh
new file mode 100755
index 000000000..d37dc5282
--- /dev/null
+++ b/benchmarks/single_node/dsv4_fp4_b200_vllm.sh
@@ -0,0 +1,120 @@
+#!/usr/bin/env bash
+
+# DeepSeek-V4-Pro B200 single-node vLLM recipe derived from the B200 pareto
+# sweep. TP mode (dp-attn=false) runs without expert parallel; DP mode
+# (dp-attn=true) enables expert parallel (EP_SIZE=TP value = DP size).
+
+source "$(dirname "$0")/../benchmark_lib.sh"
+
+check_env_vars \
+    MODEL \
+    TP \
+    DP_ATTENTION \
+    CONC \
+    ISL \
+    OSL \
+    MAX_MODEL_LEN \
+    RANDOM_RANGE_RATIO \
+    RESULT_FILENAME
+
+if [[ -n "$SLURM_JOB_ID" ]]; then
+  echo "JOB $SLURM_JOB_ID running on $SLURMD_NODENAME"
+fi
+
+nvidia-smi
+
+hf download "$MODEL"
+
+SERVER_LOG=/workspace/server.log
+PORT=${PORT:-8888}
+
+# DeepSeek-V4-Pro weights are large; engine startup can exceed the default
+# 600s. Give it an hour to load.
+export VLLM_ENGINE_READY_TIMEOUT_S=3600
+
+PARALLEL_ARGS=(--tensor-parallel-size "$TP" --data-parallel-size 1)
+if [ "${DP_ATTENTION}" = "true" ]; then
+    PARALLEL_ARGS=(--tensor-parallel-size 1 --data-parallel-size "$TP")
+fi
+
+EP_ARGS=()
+if [ "${EP_SIZE:-1}" -gt 1 ]; then
+    EP_ARGS=(--enable-expert-parallel)
+fi
+
+GMU_ARGS=()
+if [ "${DP_ATTENTION}" = "true" ]; then
+    GMU_ARGS=(--gpu-memory-utilization 0.85)
+fi
+
+if [ "${ISL}" -eq 8192 ] && [ "${CONC}" -le 128 ]; then
+    MAX_NUM_BATCHED_TOKENS=${ISL}
+else
+    MAX_NUM_BATCHED_TOKENS=2048
+fi
+
+BENCHMARK_MAX_MODEL_LEN="$MAX_MODEL_LEN"
+if [ "$ISL" -eq 1024 ] && [ "$OSL" -eq 1024 ]; then
+    BENCHMARK_MAX_MODEL_LEN=4096
+fi
+
+if [ "${EVAL_ONLY}" = "true" ]; then
+    EVAL_MAX_MODEL_LEN=$(compute_eval_context_length "$MODEL" "$BENCHMARK_MAX_MODEL_LEN")
+    export EVAL_MAX_MODEL_LEN
+    SERVE_MAX_MODEL_LEN="$EVAL_MAX_MODEL_LEN"
+else
+    SERVE_MAX_MODEL_LEN="$BENCHMARK_MAX_MODEL_LEN"
+fi
+
+# Start GPU monitoring (power, temperature, clocks every second)
+start_gpu_monitor
+
+set -x
+vllm serve "$MODEL" --host 0.0.0.0 --port "$PORT" \
+    "${PARALLEL_ARGS[@]}" \
+    --pipeline-parallel-size 1 \
+    --kv-cache-dtype fp8 \
+    --trust-remote-code \
+    --block-size 256 \
+    --no-enable-prefix-caching \
+    "${EP_ARGS[@]}" \
+    "${GMU_ARGS[@]}" \
+    --compilation-config '{"cudagraph_mode":"FULL_AND_PIECEWISE","custom_ops":["all"]}' \
+    --attention_config.use_fp4_indexer_cache True \
+    --tokenizer-mode deepseek_v4 \
+    --tool-call-parser deepseek_v4 \
+    --enable-auto-tool-choice \
+    --reasoning-parser deepseek_v4 \
+    --max-cudagraph-capture-size 2048 \
+    --max-model-len "$SERVE_MAX_MODEL_LEN" \
+    --max-num-batched-tokens "$MAX_NUM_BATCHED_TOKENS" > "$SERVER_LOG" 2>&1 &
+
+SERVER_PID=$!
+
+# Wait for server to be ready
+wait_for_server_ready --port "$PORT" --server-log "$SERVER_LOG" --server-pid "$SERVER_PID"
+
+pip install -q datasets pandas
+
+run_benchmark_serving \
+    --model "$MODEL" \
+    --port "$PORT" \
+    --backend vllm \
+    --input-len "$ISL" \
+    --output-len "$OSL" \
+    --random-range-ratio "$RANDOM_RANGE_RATIO" \
+    --num-prompts "$((CONC * 10))" \
+    --max-concurrency "$CONC" \
+    --result-filename "$RESULT_FILENAME" \
+    --result-dir /workspace/ \
+    --trust-remote-code
+
+# After throughput, run evaluation only if RUN_EVAL is true
+if [ "${RUN_EVAL}" = "true" ]; then
+    run_eval --framework lm-eval --port "$PORT"
+    append_lm_eval_summary
+fi
+
+# Stop GPU monitoring
+stop_gpu_monitor
+set +x
diff --git a/perf-changelog.yaml b/perf-changelog.yaml
index 09a791e46..22b6743b5 100644
--- a/perf-changelog.yaml
+++ b/perf-changelog.yaml
@@ -1898,3 +1898,12 @@
   description:
     - Add low-latency configs and remove non-pareto configs
   pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1193
+
+- config-keys:
+    - dsv4-fp4-b200-vllm
+  description:
+    - "Add DeepSeek-V4-Pro single-node B200 vLLM benchmark derived from B200 pareto sweep"
+    - "ISL=1024: TP8 conc 4-128; DP8 (dp-attn) conc 256-4096"
+    - "ISL=8192: TP8 conc 4-32; DP8 (dp-attn) conc 64-1024"
+  pr-link: https://github.com/SemiAnalysisAI/InferenceX/pull/1156
+
diff --git a/runners/launch_b200-cw.sh b/runners/launch_b200-cw.sh
index ec7ba9a97..0b2dbf305 100644
--- a/runners/launch_b200-cw.sh
+++ b/runners/launch_b200-cw.sh
@@ -6,6 +6,14 @@ export PORT=8888
 MODEL_CODE="${EXP_NAME%%_*}"
 FRAMEWORK_SUFFIX=$([[ "$FRAMEWORK" == "trt" ]] && printf '_trt' || printf '')
 SPEC_SUFFIX=$([[ "$SPEC_DECODING" == "mtp" ]] && printf '_mtp' || printf '')
+# Prefer a framework-tagged script (e.g. dsv4_fp4_b200_vllm.sh) so models
+# with multiple inference engines can coexist; fall back to the historical
+# name without an engine suffix (`_trt` for trt, bare for everyone else).
+BENCH_BASE="benchmarks/single_node/${MODEL_CODE}_${PRECISION}_b200"
+BENCH_SCRIPT="${BENCH_BASE}_${FRAMEWORK}${SPEC_SUFFIX}.sh"
+if [[ ! -f "$BENCH_SCRIPT" ]]; then
+    BENCH_SCRIPT="${BENCH_BASE}${FRAMEWORK_SUFFIX}${SPEC_SUFFIX}.sh"
+fi
 
 PARTITION="b200"
 SQUASH_FILE="/tmp/gharunner/squash/$(echo "$IMAGE" | sed 's/[\/:@#]/_/g').sqsh"
@@ -58,6 +66,6 @@ srun --jobid=$JOB_ID \
 --container-mount-home \
 --container-workdir=$CONTAINER_MOUNT_DIR \
 --no-container-entrypoint --export=ALL \
-bash benchmarks/single_node/${MODEL_CODE}_${PRECISION}_b200${FRAMEWORK_SUFFIX}${SPEC_SUFFIX}.sh
+bash "$BENCH_SCRIPT"
 
 scancel $JOB_ID
diff --git a/runners/launch_b200-dgxc.sh b/runners/launch_b200-dgxc.sh
index c07037ff4..edf5db957 100644
--- a/runners/launch_b200-dgxc.sh
+++ b/runners/launch_b200-dgxc.sh
@@ -253,6 +253,14 @@ else
     SQUASH_FILE="/home/sa-shared/containers/$(echo "$IMAGE" | sed 's/[\/:@#]/_/g').sqsh"
     FRAMEWORK_SUFFIX=$([[ "$FRAMEWORK" == "trt" ]] && printf '_trt' || printf '')
     SPEC_SUFFIX=$([[ "$SPEC_DECODING" == "mtp" ]] && printf '_mtp' || printf '')
+    # Prefer a framework-tagged script (e.g. dsv4_fp4_b200_vllm.sh) so models
+    # with multiple inference engines can coexist; fall back to the historical
+    # name without an engine suffix (`_trt` for trt, bare for everyone else).
+    BENCH_BASE="benchmarks/single_node/${EXP_NAME%%_*}_${PRECISION}_b200"
+    BENCH_SCRIPT="${BENCH_BASE}_${FRAMEWORK}${SPEC_SUFFIX}.sh"
+    if [[ ! -f "$BENCH_SCRIPT" ]]; then
+        BENCH_SCRIPT="${BENCH_BASE}${FRAMEWORK_SUFFIX}${SPEC_SUFFIX}.sh"
+    fi
     LOCK_FILE="${SQUASH_FILE}.lock"
 
     # TODO(Cam): lmsysorg/sglang:deepseek-v4-blackwell installs sglang editable at
@@ -290,5 +298,5 @@ else
         --no-container-mount-home \
         --container-workdir=$CONTAINER_MOUNT_DIR \
         --no-container-entrypoint --export=ALL,PORT=8888 \
-        bash benchmarks/single_node/${EXP_NAME%%_*}_${PRECISION}_b200${FRAMEWORK_SUFFIX}${SPEC_SUFFIX}.sh
+        bash "$BENCH_SCRIPT"
 fi
diff --git a/runners/launch_b200-nb.sh b/runners/launch_b200-nb.sh
index 6b411fec2..e0c8d92fb 100644
--- a/runners/launch_b200-nb.sh
+++ b/runners/launch_b200-nb.sh
@@ -4,6 +4,14 @@ HF_HUB_CACHE_MOUNT="/mnt/data/gharunners/hf-hub-cache/"
 PARTITION="main"
 FRAMEWORK_SUFFIX=$([[ "$FRAMEWORK" == "trt" ]] && printf '_trt' || printf '')
 SPEC_SUFFIX=$([[ "$SPEC_DECODING" == "mtp" ]] && printf '_mtp' || printf '')
+# Prefer a framework-tagged script (e.g. dsv4_fp4_b200_vllm.sh) so models
+# with multiple inference engines can coexist; fall back to the historical
+# name without an engine suffix (`_trt` for trt, bare for everyone else).
+BENCH_BASE="benchmarks/single_node/${EXP_NAME%%_*}_${PRECISION}_b200"
+BENCH_SCRIPT="${BENCH_BASE}_${FRAMEWORK}${SPEC_SUFFIX}.sh"
+if [[ ! -f "$BENCH_SCRIPT" ]]; then
+    BENCH_SCRIPT="${BENCH_BASE}${FRAMEWORK_SUFFIX}${SPEC_SUFFIX}.sh"
+fi
 
 UCX_NET_DEVICES=eth0
 
@@ -27,4 +35,4 @@ srun --partition=$PARTITION --gres=gpu:$TP --exclusive --job-name="$RUNNER_NAME"
 --container-writable \
 --container-workdir=$CONTAINER_MOUNT_DIR \
 --no-container-entrypoint --export=ALL,PORT=8888,UCX_NET_DEVICES=$UCX_NET_DEVICES \
-bash benchmarks/single_node/${EXP_NAME%%_*}_${PRECISION}_b200${FRAMEWORK_SUFFIX}${SPEC_SUFFIX}.sh
\ No newline at end of file
+bash "$BENCH_SCRIPT"
\ No newline at end of file