diff --git a/.github/workflows/scripts/build.sh b/.github/workflows/scripts/build.sh
index 5bb9f719680..eb3971c42bf 100644
--- a/.github/workflows/scripts/build.sh
+++ b/.github/workflows/scripts/build.sh
@@ -9,6 +9,9 @@ PATH=${cuda_home}/bin:$PATH
 LD_LIBRARY_PATH=${cuda_home}/lib64:$LD_LIBRARY_PATH
 
 # Install requirements
+if [ "$(echo $2 | cut -d. -f1)" = "12" ]; then
+    sed -i 's/^nvidia-cutlass-dsl\[cu13\]>=/nvidia-cutlass-dsl>=/' requirements/cuda.txt
+fi
 $python_executable -m pip install -r requirements/build/cuda.txt -r requirements/cuda.txt
 
 # Limit the number of parallel jobs to avoid OOM
diff --git a/docker/Dockerfile b/docker/Dockerfile
index 7d8decfe870..54381512cca 100644
--- a/docker/Dockerfile
+++ b/docker/Dockerfile
@@ -199,7 +199,10 @@ COPY requirements/cuda.txt requirements/cuda.txt
 COPY use_existing_torch.py use_existing_torch.py
 COPY pyproject.toml pyproject.toml
 RUN --mount=type=cache,target=/root/.cache/uv \
-    if [ "${PYTORCH_NIGHTLY}" = "1" ]; then \
+    if [ "$(echo $CUDA_VERSION | cut -d. -f1)" = "12" ]; then \
+        sed -i 's/^nvidia-cutlass-dsl\[cu13\]>=/nvidia-cutlass-dsl>=/' requirements/cuda.txt; \
+    fi \
+    && if [ "${PYTORCH_NIGHTLY}" = "1" ]; then \
         echo "Installing torch nightly..." \
         && uv pip install --python /opt/venv/bin/python3 torch torchaudio torchvision --pre \
         --index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/nightly/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.') \
@@ -626,6 +629,9 @@ ARG PYTORCH_CUDA_INDEX_BASE_URL
 COPY requirements/common.txt /tmp/common.txt
 COPY requirements/cuda.txt /tmp/requirements-cuda.txt
 RUN --mount=type=cache,target=/root/.cache/uv \
+    if [ "$(echo $CUDA_VERSION | cut -d. -f1)" = "12" ]; then \
+        sed -i 's/^nvidia-cutlass-dsl\[cu13\]>=/nvidia-cutlass-dsl>=/' /tmp/requirements-cuda.txt; \
+    fi && \
     uv pip install --system -r /tmp/requirements-cuda.txt \
         --extra-index-url ${PYTORCH_CUDA_INDEX_BASE_URL}/cu$(echo $CUDA_VERSION | cut -d. -f1,2 | tr -d '.') && \
     rm /tmp/requirements-cuda.txt /tmp/common.txt
diff --git a/requirements/cuda.txt b/requirements/cuda.txt
index abff2525af9..068e83272a5 100644
--- a/requirements/cuda.txt
+++ b/requirements/cuda.txt
@@ -21,5 +21,5 @@ nvidia-cudnn-frontend>=1.13.0,<1.19.0
 fastsafetensors >= 0.2.2
 
 # QuACK and Cutlass DSL for FA4 (cute-DSL implementation)
-nvidia-cutlass-dsl>=4.4.2
+nvidia-cutlass-dsl[cu13]>=4.4.2
 quack-kernels>=0.3.3
diff --git a/setup.py b/setup.py
index dc963c9e789..738a89a4775 100644
--- a/setup.py
+++ b/setup.py
@@ -970,6 +970,9 @@ def get_requirements() -> list[str]:
                 # vllm-flash-attn is built only for CUDA 12.x.
                 # Skip for other versions.
                 continue
+            if "nvidia-cutlass-dsl[cu13]" in req and cuda_major == "12":
+                # [cu13] extra is the default; strip it on CUDA 12 builds.
+                req = req.replace("nvidia-cutlass-dsl[cu13]", "nvidia-cutlass-dsl")
             modified_requirements.append(req)
         requirements = modified_requirements
     elif _is_hip():