diff --git a/.github/workflows/rel_sgl-kernel-xpu.yml b/.github/workflows/rel_sgl-kernel-xpu.yml
new file mode 100644
index 000000000..c5a433298
--- /dev/null
+++ b/.github/workflows/rel_sgl-kernel-xpu.yml
@@ -0,0 +1,86 @@
+name: Release Docker Images (sgl-kernel-xpu rel)
+on:
+  workflow_dispatch:
+    inputs:
+      wheel_url:
+        description: 'S3 URL of the pre-built sgl-kernel-xpu wheel to bake into the image'
+        required: true
+        default: 'https://sgl-kernel-xpu.s3.us-east-1.amazonaws.com/wheels/release-v0.1.0/sglang_kernel_xpu-0.1.0-cp310-abi3-manylinux_2_38_x86_64.whl'
+      version:
+        description: 'Release version tag (e.g. v0.1.0). Used as part of the image tag.'
+        required: true
+        default: 'v0.1.0'
+  push:
+    tags:
+      - 'v*'
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.number || github.ref }}
+  cancel-in-progress: true
+
+jobs:
+  publish:
+    if: github.repository == 'sgl-project/sgl-kernel-xpu'
+    runs-on: sglang-bmg
+    environment: 'prod'
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@v4
+        with:
+          path: sgl-kernel-xpu-rel-build
+          fetch-depth: 0
+
+      - name: Resolve release metadata
+        id: image_meta
+        run: |
+          cd sgl-kernel-xpu-rel-build
+          commit_hash=$(git rev-parse --short=7 HEAD)
+          if [ -z "${commit_hash}" ]; then
+            echo "::error::Could not resolve commit hash"
+            exit 1
+          fi
+          if [ "${GITHUB_EVENT_NAME}" = "push" ]; then
+            version="${GITHUB_REF_NAME}"
+            wheel_url="https://sgl-kernel-xpu.s3.us-east-1.amazonaws.com/wheels/release-${version}/sglang_kernel_xpu-${version#v}-cp310-abi3-manylinux_2_38_x86_64.whl"
+          else
+            version="${{ inputs.version }}"
+            wheel_url="${{ inputs.wheel_url }}"
+          fi
+          image_tag="${version}-${commit_hash}"
+          echo "commit_hash=${commit_hash}" >> $GITHUB_OUTPUT
+          echo "IMAGE_TAG=${image_tag}" >> $GITHUB_ENV
+          echo "WHEEL_URL=${wheel_url}" >> $GITHUB_ENV
+          echo "Image tag: ${image_tag}"
+          echo "Wheel URL: ${wheel_url}"
+
+      - name: Set up Docker Buildx
+        uses: docker/setup-buildx-action@v3
+
+      - name: Login to Docker Hub (Intel)
+        uses: docker/login-action@v3
+        with:
+          username: ${{ secrets.DOCKERHUB_INTEL_USERNAME }}
+          password: ${{ secrets.DOCKERHUB_INTEL_TOKEN }}
+
+      - name: Build intel/sgl-kernel-xpu-rel
+        run: |
+          echo "Building intel/sgl-kernel-xpu-rel:${{ env.IMAGE_TAG }} from ${{ env.WHEEL_URL }}"
+          docker build sgl-kernel-xpu-rel-build \
+            -f sgl-kernel-xpu-rel-build/Dockerfile.rel_xpu_kernel \
+            --build-arg WHEEL_URL="${{ env.WHEEL_URL }}" \
+            --no-cache --progress=plain \
+            -t "intel/sgl-kernel-xpu-rel:${{ env.IMAGE_TAG }}" \
+            -t "intel/sgl-kernel-xpu-rel:rel"
+
+      - name: Push intel/sgl-kernel-xpu-rel
+        run: |
+          push_with_retry() {
+            for i in 1 2 3 4 5; do
+              docker push "$1" && return 0
+              echo "push failed (attempt $i), retrying in $((i*15))s..."
+              sleep $((i*15))
+            done
+            return 1
+          }
+          push_with_retry "intel/sgl-kernel-xpu-rel:${{ env.IMAGE_TAG }}"
+          push_with_retry "intel/sgl-kernel-xpu-rel:rel"
diff --git a/Dockerfile.rel_xpu_kernel b/Dockerfile.rel_xpu_kernel
new file mode 100644
index 000000000..8af06b33f
--- /dev/null
+++ b/Dockerfile.rel_xpu_kernel
@@ -0,0 +1,100 @@
+# If the device is Battlemage, we need to set UBUNTU_VERSION to 24.10
+
+# Usage: docker build \
+#   --build-arg UBUNTU_VERSION=24.04 \
+#   --build-arg PYTHON_VERSION=3.10 \
+#   --build-arg WHEEL_URL=https://sgl-kernel-xpu.s3.us-east-1.amazonaws.com/wheels/release-v0.1.0/sglang_kernel_xpu-0.1.0-cp310-abi3-manylinux_2_38_x86_64.whl \
+#   -t sglang:rel_xpu_kernel -f Dockerfile.rel_xpu_kernel --no-cache .
+
+
+# Set default Ubuntu version to 24.04
+FROM intel/deep-learning-essentials:2026.0.0-devel-ubuntu24.04
+
+ENV DEBIAN_FRONTEND=noninteractive
+
+# Define build arguments
+ARG PYTHON_VERSION=3.10
+
+# Set environment variables
+
+ARG SG_LANG_REPO=https://github.com/sgl-project/sglang.git
+ARG SG_LANG_BRANCH=main
+
+# Pre-built sgl-kernel-xpu wheel uploaded to S3 by build-and-upload-wheel.yml.
+ARG WHEEL_URL=https://sgl-kernel-xpu.s3.us-east-1.amazonaws.com/wheels/release-v0.1.0/sglang_kernel_xpu-0.1.0-cp310-abi3-manylinux_2_38_x86_64.whl
+
+# Pin the compute UMD + IGC. The kobuk-team PPA is rolling; an untested build
+# mismatched the host xe KMD and faulted libze_intel_gpu.so on the B580 runner.
+# Must stay in lockstep with the host KMD; override via --build-arg per workflow.
+ARG COMPUTE_RUNTIME_VERSION=26.18.38308.1
+ARG IGC_VERSION=2.34.4+21428
+ARG GMM_VERSION=22.10.0
+RUN apt-get install -y software-properties-common && \
+    add-apt-repository -y ppa:kobuk-team/intel-graphics && \
+    apt-get update  && \
+    # Loader + media/metrics from the PPA; GPU driver itself is pinned below.
+    apt-get install -y libze1 intel-metrics-discovery clinfo intel-gsc && \
+    apt-get install -y intel-media-va-driver-non-free libmfx-gen1 libvpl2 libvpl-tools libva-glx2 va-driver-all vainfo && \
+    apt-get install -y libze-dev && \
+    # IGC first: libze-intel-gpu1 / intel-opencl-icd depend on it.
+    cd /tmp && \
+    igc_url="https://github.com/intel/intel-graphics-compiler/releases/download/v${IGC_VERSION%%+*}" && \
+    cr_url="https://github.com/intel/compute-runtime/releases/download/${COMPUTE_RUNTIME_VERSION}" && \
+    curl -fsSL -O "${igc_url}/intel-igc-core-2_${IGC_VERSION}_amd64.deb" && \
+    curl -fsSL -O "${igc_url}/intel-igc-opencl-2_${IGC_VERSION}_amd64.deb" && \
+    curl -fsSL -O "${cr_url}/libze-intel-gpu1_${COMPUTE_RUNTIME_VERSION}-0_amd64.deb" && \
+    curl -fsSL -O "${cr_url}/intel-opencl-icd_${COMPUTE_RUNTIME_VERSION}-0_amd64.deb" && \
+    curl -fsSL -O "${cr_url}/intel-ocloc_${COMPUTE_RUNTIME_VERSION}-0_amd64.deb" && \
+    curl -fsSL -O "${cr_url}/libigdgmm12_${GMM_VERSION}_amd64.deb" && \
+    apt-get install -y --allow-downgrades \
+        ./intel-igc-core-2_${IGC_VERSION}_amd64.deb \
+        ./intel-igc-opencl-2_${IGC_VERSION}_amd64.deb \
+        ./libigdgmm12_${GMM_VERSION}_amd64.deb \
+        ./libze-intel-gpu1_${COMPUTE_RUNTIME_VERSION}-0_amd64.deb \
+        ./intel-opencl-icd_${COMPUTE_RUNTIME_VERSION}-0_amd64.deb \
+        ./intel-ocloc_${COMPUTE_RUNTIME_VERSION}-0_amd64.deb && \
+    rm -f /tmp/*.deb && \
+    # Hold so a later apt upgrade cannot pull the rolling PPA version.
+    apt-mark hold libze-intel-gpu1 intel-opencl-icd intel-ocloc libigdgmm12 \
+        intel-igc-core-2 intel-igc-opencl-2
+
+# Install Miniforge & PyTorch/Triton
+RUN curl -fsSL -v -o miniforge.sh -O https://github.com/conda-forge/miniforge/releases/download/25.1.1-0/Miniforge3-Linux-x86_64.sh && \
+    bash miniforge.sh -b -p ./miniforge3 && \
+    rm miniforge.sh && \
+    . ./miniforge3/bin/activate && \
+    conda create -y -n py${PYTHON_VERSION} python=${PYTHON_VERSION} && conda activate py${PYTHON_VERSION} && \
+    conda install pip && \
+    echo ". /miniforge3/bin/activate; conda activate py${PYTHON_VERSION}; . /opt/intel/oneapi/setvars.sh; cd /root/" >> /root/.bashrc;
+
+RUN --mount=type=secret,id=github_token \
+    cd /root && \
+    . /miniforge3/bin/activate && \
+    conda activate py${PYTHON_VERSION} && \
+    # . /opt/intel/oneapi/setvars.sh --force && \
+    # Install Torch
+    pip install torch==2.13.0 torchvision triton-xpu --index-url https://download.pytorch.org/whl/xpu
+
+# Install SGlang from source; install sgl-kernel-xpu from the pre-built wheel on S3.
+RUN --mount=type=secret,id=github_token \
+    cd /root && \
+    . /miniforge3/bin/activate && \
+    conda activate py${PYTHON_VERSION} && \
+    echo "cloning ${SG_LANG_BRANCH} from ${SG_LANG_REPO}" && \
+    git clone --branch ${SG_LANG_BRANCH} --single-branch ${SG_LANG_REPO} && \
+    cd sglang && \
+    # Download the release wheel from S3 and install it.
+    echo "fetching sgl-kernel-xpu wheel from ${WHEEL_URL}" && \
+    mkdir -p /root/rel-wheel && \
+    curl -fSL --retry 5 --retry-delay 10 -o "/root/rel-wheel/$(basename "${WHEEL_URL%%\?*}")" "${WHEEL_URL}" && \
+    ls -la /root/rel-wheel/ && \
+    pip install "/root/rel-wheel/$(basename "${WHEEL_URL%%\?*}")" && \
+    # Install required packages for sglang workloads
+    pip install msgspec blake3 py-cpuinfo compressed_tensors gguf partial_json_parser einops matplotlib pandas --root-user-action=ignore aiohttp && \
+    conda install libsqlite=3.48.0 -y && \
+    echo ". /miniforge3/bin/activate; conda activate py${PYTHON_VERSION}; cd /root/" >> /root/.bashrc;
+
+# Set the default shell to bash
+SHELL ["bash", "-c"]
+CMD ["bash", "-c", "source /root/.bashrc && exec bash"]
+USER root