diff --git a/tools/launcher/common/specdec_bench/_cells/qwen35_4b_mtp_vllm_t1_d3.yaml b/tools/launcher/common/specdec_bench/_cells/qwen35_4b_mtp_vllm_t1_d3.yaml
new file mode 100644
index 00000000000..2c3bb178e21
--- /dev/null
+++ b/tools/launcher/common/specdec_bench/_cells/qwen35_4b_mtp_vllm_t1_d3.yaml
@@ -0,0 +1,4 @@
+sampling_kwargs:
+  temperature: 1
+engine_args:
+  max_model_len: 40960
diff --git a/tools/launcher/examples/Qwen/Qwen3.5-4B/specdec_bench_mtp_vllm_t1_d3.yaml b/tools/launcher/examples/Qwen/Qwen3.5-4B/specdec_bench_mtp_vllm_t1_d3.yaml
new file mode 100644
index 00000000000..881e8771ecb
--- /dev/null
+++ b/tools/launcher/examples/Qwen/Qwen3.5-4B/specdec_bench_mtp_vllm_t1_d3.yaml
@@ -0,0 +1,66 @@
+# SPEED-bench MTP speculative-decoding run for Qwen3.5-4B via vLLM.
+#
+# The qwen3_5 model_type needs transformers >= 4.58, which is not in
+# vllm/vllm-openai:latest yet — use the qwen3_5-cu130 tag instead.
+#
+# Slurm run on cw_dfw:
+#   uv run slurm.py --yaml modules/Model-Optimizer/tools/launcher/examples/Qwen/Qwen3.5-4B/specdec_bench_mtp_vllm_t1_d3.yaml --yes
+
+job_name: Qwen3.5-4B_specdec_bench_mtp_vllm_t1_d3
+
+pipeline:
+  global_vars:
+    hf_model: /hf-local/Qwen/Qwen3.5-4B
+
+  task_0:
+    script: common/specdec_bench/run.sh
+    args:
+      - --dataset speed
+      - --dataset_path /hf-local/nvidia/SPEED-Bench-Internal/qualitative
+      - --engine VLLM
+      - --speculative_algorithm MTP
+      - --draft_length 3
+      - --runtime_params common/specdec_bench/_cells/qwen35_4b_mtp_vllm_t1_d3.yaml
+      - --tp_size 2
+      - --ep_size 1
+      - --concurrency 32
+      - --output_length 4096
+      - --aa_timing
+      - --show_progress
+      - --save_dir /scratchspace/qwen35_4b_mtp_vllm_t1_d3/qualitative
+    environment:
+      - HF_MODEL_CKPT: <<global_vars.hf_model>>
+      - HF_LOCAL: /hf-local
+    slurm_config:
+      _factory_: "slurm_factory"
+      nodes: 1
+      ntasks_per_node: 1
+      gpus_per_node: 2
+      container: vllm/vllm-openai:qwen3_5-cu130
+
+  task_1:
+    script: common/specdec_bench/run.sh
+    args:
+      - --dataset speed
+      - --dataset_path /hf-local/nvidia/SPEED-Bench-Internal/throughput_32k
+      - --engine VLLM
+      - --speculative_algorithm MTP
+      - --draft_length 3
+      - --runtime_params common/specdec_bench/_cells/qwen35_4b_mtp_vllm_t1_d3.yaml
+      - --tp_size 2
+      - --ep_size 1
+      - --concurrency 8
+      - --num_requests 80
+      - --output_length 4096
+      - --aa_timing
+      - --show_progress
+      - --save_dir /scratchspace/qwen35_4b_mtp_vllm_t1_d3/throughput_32k
+    environment:
+      - HF_MODEL_CKPT: <<global_vars.hf_model>>
+      - HF_LOCAL: /hf-local
+    slurm_config:
+      _factory_: "slurm_factory"
+      nodes: 1
+      ntasks_per_node: 1
+      gpus_per_node: 2
+      container: vllm/vllm-openai:qwen3_5-cu130