diff --git a/.github/workflows/nightly-test-npu.yml b/.github/workflows/nightly-test-npu.yml
index 1ab6b6733..452dfcade 100644
--- a/.github/workflows/nightly-test-npu.yml
+++ b/.github/workflows/nightly-test-npu.yml
@@ -35,7 +35,7 @@ jobs:
       matrix:
         part: [0, 1]
     container:
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc2-a3-ubuntu22.04-py3.11
+      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-a3-ubuntu22.04-py3.11
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
@@ -70,7 +70,6 @@ jobs:
           PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
           STREAMS_PER_DEVICE: 32
         run: |
-          export PATH="/usr/local/Ascend/8.3.RC1/compiler/bishengir/bin:${PATH}"
           pip install sentence_transformers accelerate
           cd test
           python3 run_suite.py --hw npu --suite nightly-1-npu-a3 --nightly --continue-on-error --timeout-per-file 3600 --auto-partition-id ${{ matrix.part }} --auto-partition-size 2
@@ -83,7 +82,7 @@ jobs:
       matrix:
         part: [0]
     container:
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc2-a3-ubuntu22.04-py3.11
+      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-a3-ubuntu22.04-py3.11
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
@@ -118,7 +117,6 @@ jobs:
           PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
           STREAMS_PER_DEVICE: 32
         run: |
-          export PATH="/usr/local/Ascend/8.3.RC1/compiler/bishengir/bin:${PATH}"
           pip install sentence_transformers accelerate
           cd test
           python3 run_suite.py --hw npu --suite nightly-2-npu-a3 --nightly --continue-on-error --timeout-per-file 3600 --auto-partition-id ${{ matrix.part }} --auto-partition-size 1
@@ -131,7 +129,7 @@ jobs:
       matrix:
         part: [0]
     container:
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc2-a3-ubuntu22.04-py3.11
+      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-a3-ubuntu22.04-py3.11
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
@@ -144,8 +142,7 @@ jobs:
           CACHING_URL="cache-service.nginx-pypi-cache.svc.cluster.local"
           sed -Ei "s@(ports|archive).ubuntu.com@${CACHING_URL}:8081@g" /etc/apt/sources.list
           pip config set global.index-url http://${CACHING_URL}/pypi/simple
-          pip config set global.extra-index-url "https://pypi.tuna.tsinghua.edu.cn/simple"
-          pip config set global.trusted-host "${CACHING_URL} pypi.tuna.tsinghua.edu.cn"
+          pip config set global.trusted-host "${CACHING_URL}"
 
           bash scripts/ci/npu/npu_ci_install_dependency.sh a3
           # copy required file from our daily cache
@@ -167,7 +164,6 @@ jobs:
           PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
           STREAMS_PER_DEVICE: 32
         run: |
-          export PATH="/usr/local/Ascend/8.3.RC1/compiler/bishengir/bin:${PATH}"
           hf download lmms-lab/MMMU --repo-type dataset
           pip install sentence_transformers torchaudio==2.8.0 torch_npu==2.8.0
           pip install protobuf==6.31.1 zss pre-commit wandb>=0.16.0 tenacity==8.3.0 loguru openpyxl latex2sympy2 zstandard transformers-stream-generator tqdm-multiprocess pycocoevalcap
diff --git a/.github/workflows/pr-test-npu.yml b/.github/workflows/pr-test-npu.yml
index de7833571..0361d078f 100644
--- a/.github/workflows/pr-test-npu.yml
+++ b/.github/workflows/pr-test-npu.yml
@@ -79,7 +79,7 @@ jobs:
     if: needs.check-changes.outputs.main_package == 'true'
     runs-on: linux-aarch64-a2-1
     container:
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc2-910b-ubuntu22.04-py3.11
+      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-910b-ubuntu22.04-py3.11
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
@@ -111,7 +111,6 @@ jobs:
           PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
           STREAMS_PER_DEVICE: 32
         run: |
-          export PATH="/usr/local/Ascend/8.3.RC1/compiler/bishengir/bin:${PATH}"
           cd test/srt
           python3 run_suite.py --suite per-commit-1-npu-a2
 
@@ -124,7 +123,7 @@ jobs:
       matrix:
         part: [0, 1, 2]
     container:
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc2-910b-ubuntu22.04-py3.11
+      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-910b-ubuntu22.04-py3.11
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
@@ -156,7 +155,6 @@ jobs:
           PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
           STREAMS_PER_DEVICE: 32
         run: |
-          export PATH="/usr/local/Ascend/8.3.RC1/compiler/bishengir/bin:${PATH}"
           cd test/srt
           python3 run_suite.py --suite per-commit-2-npu-a2 --auto-partition-id ${{ matrix.part }} --auto-partition-size 3
 
@@ -165,7 +163,7 @@ jobs:
     if: needs.check-changes.outputs.main_package == 'true'
     runs-on: linux-aarch64-a2-4
     container:
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc2-910b-ubuntu22.04-py3.11
+      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-910b-ubuntu22.04-py3.11
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
@@ -197,7 +195,6 @@ jobs:
           PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
           STREAMS_PER_DEVICE: 32
         run: |
-          export PATH="/usr/local/Ascend/8.3.RC1/compiler/bishengir/bin:${PATH}"
           cd test/srt
           python3 run_suite.py --suite per-commit-4-npu-a2 --timeout-per-file 3600
 
@@ -210,7 +207,7 @@ jobs:
       matrix:
         part: [0, 1]
     container:
-      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.3.rc2-a3-ubuntu22.04-py3.11
+      image: swr.cn-southwest-2.myhuaweicloud.com/base_image/ascend-ci/cann:8.5.0-a3-ubuntu22.04-py3.11
     steps:
       - name: Checkout code
         uses: actions/checkout@v4
@@ -242,6 +239,5 @@ jobs:
           PYTORCH_NPU_ALLOC_CONF: "expandable_segments:True"
           STREAMS_PER_DEVICE: 32
         run: |
-          export PATH="/usr/local/Ascend/8.3.RC1/compiler/bishengir/bin:${PATH}"
           cd test/srt
           python3 run_suite.py --suite per-commit-16-npu-a3 --timeout-per-file 3600 --auto-partition-id ${{ matrix.part }} --auto-partition-size 2
diff --git a/.github/workflows/release-docker-npu-nightly.yml b/.github/workflows/release-docker-npu-nightly.yml
index 7b66eba24..e798bfa0f 100644
--- a/.github/workflows/release-docker-npu-nightly.yml
+++ b/.github/workflows/release-docker-npu-nightly.yml
@@ -13,7 +13,7 @@ jobs:
     runs-on: ubuntu-22.04-arm
     strategy:
       matrix:
-        cann_version: ["8.3.rc2"]
+        cann_version: ["8.5.0"]
         device_type: ["910b", "a3"]
     steps:
       - name: Checkout repository
@@ -67,6 +67,6 @@ jobs:
           push: ${{ github.repository == 'sgl-project/sglang' && github.event_name != 'pull_request' }}
           provenance: false
           build-args: |
-            SGLANG_KERNEL_NPU_TAG=2025.12.31
+            SGLANG_KERNEL_NPU_TAG=2026.01.21
             CANN_VERSION=${{ matrix.cann_version }}
             DEVICE_TYPE=${{ matrix.device_type }}
diff --git a/.github/workflows/release-docker-npu.yml b/.github/workflows/release-docker-npu.yml
index 850efbae0..2bd8f1490 100644
--- a/.github/workflows/release-docker-npu.yml
+++ b/.github/workflows/release-docker-npu.yml
@@ -14,7 +14,7 @@ jobs:
     runs-on: ubuntu-22.04-arm
     strategy:
       matrix:
-        cann_version: ["8.3.rc2"]
+        cann_version: ["8.5.0"]
         device_type: ["910b", "a3"]
     steps:
       - name: Checkout repository
@@ -81,7 +81,7 @@ jobs:
           push: ${{ github.repository == 'sgl-project/sglang' && github.event_name != 'pull_request' }}
           provenance: false
           build-args: |
-            SGLANG_KERNEL_NPU_TAG=2025.12.31
+            SGLANG_KERNEL_NPU_TAG=2026.01.21
             CANN_VERSION=${{ matrix.cann_version }}
             DEVICE_TYPE=${{ matrix.device_type }}
             SGLANG_TAG=${{ steps.version.outputs.version }}
diff --git a/docker/npu.Dockerfile b/docker/npu.Dockerfile
index e49551b19..8f5c065e9 100644
--- a/docker/npu.Dockerfile
+++ b/docker/npu.Dockerfile
@@ -1,4 +1,4 @@
-ARG CANN_VERSION=8.3.rc2
+ARG CANN_VERSION=8.5.0
 ARG DEVICE_TYPE=a3
 ARG OS=ubuntu22.04
 ARG PYTHON_VERSION=py3.11
@@ -6,14 +6,13 @@ ARG PYTHON_VERSION=py3.11
 FROM quay.io/ascend/cann:$CANN_VERSION-$DEVICE_TYPE-$OS-$PYTHON_VERSION
 
 # Update pip & apt sources
+ARG CANN_VERSION
+ARG DEVICE_TYPE
 ARG PIP_INDEX_URL="https://pypi.org/simple/"
 ARG APTMIRROR=""
 ARG PYTORCH_VERSION="2.8.0"
 ARG TORCHVISION_VERSION="0.23.0"
-ARG PTA_URL="https://sglang-ascend.obs.cn-east-3.myhuaweicloud.com/sglang/torch_npu/torch_npu-2.8.0.post2.dev20251113-cp311-cp311-manylinux_2_28_aarch64.whl"
-ARG TRITON_ASCEND_URL="https://sglang-ascend.obs.cn-east-3.myhuaweicloud.com/sglang/triton_ascend/triton_ascend-3.2.0.dev2025112116-cp311-cp311-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl"
-ARG BISHENG_NAME="Ascend-BiSheng-toolkit_aarch64_20251121.run"
-ARG BISHENG_URL="https://sglang-ascend.obs.cn-east-3.myhuaweicloud.com/sglang/triton_ascend/${BISHENG_NAME}"
+ARG PTA_URL="https://gitcode.com/Ascend/pytorch/releases/download/v7.3.0-pytorch2.8.0/torch_npu-2.8.0.post2-cp311-cp311-manylinux_2_28_aarch64.whl"
 ARG SGLANG_TAG=main
 ARG ASCEND_CANN_PATH=/usr/local/Ascend/ascend-toolkit
 ARG SGLANG_KERNEL_NPU_TAG=main
@@ -31,6 +30,7 @@ RUN if [ -n "$APTMIRROR" ];then sed -i "s|.*.ubuntu.com|$APTMIRROR|g" /etc/apt/s
 
 # Install development tools and utilities
 RUN apt-get update -y && apt upgrade -y && apt-get install -y \
+    unzip \
     build-essential \
     cmake \
     vim \
@@ -57,7 +57,7 @@ ENV LC_ALL=en_US.UTF-8
 
 
 ### Install MemFabric
-RUN ${PIP_INSTALL} memfabric-hybrid==1.0.0
+RUN ${PIP_INSTALL} memfabric-hybrid==1.0.5
 ### Install SGLang Model Gateway
 RUN ${PIP_INSTALL} sglang-router
 
@@ -69,7 +69,7 @@ RUN (${PIP_INSTALL} torch==${PYTORCH_VERSION} torchvision==${TORCHVISION_VERSION
 
 # TODO: install from pypi released triton-ascend
 RUN (${PIP_INSTALL} pybind11) \
-    && (${PIP_INSTALL} ${TRITON_ASCEND_URL})
+    && (${PIP_INSTALL} triton-ascend)
 
 # Install SGLang
 RUN git clone https://github.com/sgl-project/sglang --branch $SGLANG_TAG && \
@@ -78,23 +78,13 @@ RUN git clone https://github.com/sgl-project/sglang --branch $SGLANG_TAG && \
 
 # Install Deep-ep
 # pin wheel to 0.45.1 ref: https://github.com/pypa/wheel/issues/662
-RUN ${PIP_INSTALL} wheel==0.45.1 && git clone --branch $SGLANG_KERNEL_NPU_TAG https://github.com/sgl-project/sgl-kernel-npu.git \
-    && export LD_LIBRARY_PATH=${ASCEND_CANN_PATH}/latest/runtime/lib64/stub:$LD_LIBRARY_PATH && \
-    source ${ASCEND_CANN_PATH}/set_env.sh && \
-    cd sgl-kernel-npu && \
-    bash build.sh \
+RUN ${PIP_INSTALL} wheel==0.45.1 pybind11 pyyaml decorator scipy attrs psutil \
+    && mkdir sgl-kernel-npu \
+    && cd sgl-kernel-npu \
+    && wget https://github.com/sgl-project/sgl-kernel-npu/releases/download/${SGLANG_KERNEL_NPU_TAG}/sgl-kernel-npu_${SGLANG_KERNEL_NPU_TAG}_${CANN_VERSION}_${DEVICE_TYPE}.zip \
+    && unzip sgl-kernel-npu_${SGLANG_KERNEL_NPU_TAG}_${CANN_VERSION}_${DEVICE_TYPE}.zip \
     && ${PIP_INSTALL} output/deep_ep*.whl output/sgl_kernel_npu*.whl \
     && cd .. && rm -rf sgl-kernel-npu \
-    && cd "$(python3 -m pip show deep-ep | awk '/^Location:/ {print $2}')" && ln -s deep_ep/deep_ep_cpp*.so
-
-# Install CustomOps
-RUN wget https://sglang-ascend.obs.cn-east-3.myhuaweicloud.com/ops/CANN-custom_ops-8.2.0.0-$DEVICE_TYPE-linux.aarch64.run && \
-    chmod a+x ./CANN-custom_ops-8.2.0.0-$DEVICE_TYPE-linux.aarch64.run && \
-    ./CANN-custom_ops-8.2.0.0-$DEVICE_TYPE-linux.aarch64.run --quiet --install-path=/usr/local/Ascend/ascend-toolkit/latest/opp && \
-    wget https://sglang-ascend.obs.cn-east-3.myhuaweicloud.com/ops/custom_ops-1.0.$DEVICE_TYPE-cp311-cp311-linux_aarch64.whl && \
-    ${PIP_INSTALL} ./custom_ops-1.0.$DEVICE_TYPE-cp311-cp311-linux_aarch64.whl
-
-# Install Bisheng
-RUN wget -O "${BISHENG_NAME}" "${BISHENG_URL}" && chmod a+x "${BISHENG_NAME}" && "./${BISHENG_NAME}" --install && rm "${BISHENG_NAME}"
+    && cd "$(python3 -m pip show deep-ep | awk '/^Location:/ {print $2}')" && ln -sf deep_ep/deep_ep_cpp*.so
 
 CMD ["/bin/bash"]
diff --git a/python/sglang/srt/hardware_backend/npu/attention/ascend_backend.py b/python/sglang/srt/hardware_backend/npu/attention/ascend_backend.py
index aae19147a..81e1241ee 100644
--- a/python/sglang/srt/hardware_backend/npu/attention/ascend_backend.py
+++ b/python/sglang/srt/hardware_backend/npu/attention/ascend_backend.py
@@ -546,7 +546,7 @@ class AscendAttnBackend(AttentionBackend):
         actual_seq_qlen_prev, actual_seq_qlen_next = actual_seq_qlen
         actual_seq_lengths_kv_prev, actual_seq_lengths_kv_next = actual_seq_lengths_kv
 
-        attn_out_prev = torch.ops.custom.npu_sparse_flash_attention(
+        attn_out_prev, _, _ = torch_npu.npu_sparse_flash_attention(
             query=q_nope_prev,
             key=k_nope,
             value=k_nope,
@@ -565,8 +565,10 @@ class AscendAttnBackend(AttentionBackend):
             layout_query="TND",
             layout_kv="PA_BSND",
             sparse_mode=3,
+            attention_mode=2,
+            return_softmax_lse=False,
         )
-        attn_out_next = torch.ops.custom.npu_sparse_flash_attention(
+        attn_out_next, _, _ = torch_npu.npu_sparse_flash_attention(
             query=q_nope_next,
             key=k_nope,
             value=k_nope,
@@ -585,6 +587,8 @@ class AscendAttnBackend(AttentionBackend):
             layout_query="TND",
             layout_kv="PA_BSND",
             sparse_mode=3,
+            attention_mode=2,
+            return_softmax_lse=False,
         )
         return torch.cat([attn_out_prev, attn_out_next], dim=0)
 
@@ -675,7 +679,7 @@ class AscendAttnBackend(AttentionBackend):
                 actual_seq_lengths_kv,
             )
         else:
-            attn_out = torch.ops.custom.npu_sparse_flash_attention(
+            attn_out, _, _ = torch_npu.npu_sparse_flash_attention(
                 query=q_nope,
                 key=k_nope,
                 value=k_nope,
@@ -694,6 +698,8 @@ class AscendAttnBackend(AttentionBackend):
                 layout_query="TND",
                 layout_kv="PA_BSND",
                 sparse_mode=3,
+                attention_mode=2,
+                return_softmax_lse=False,
             )
 
         return attn_out
diff --git a/python/sglang/srt/hardware_backend/npu/utils.py b/python/sglang/srt/hardware_backend/npu/utils.py
index 478c73d24..3e09841bd 100644
--- a/python/sglang/srt/hardware_backend/npu/utils.py
+++ b/python/sglang/srt/hardware_backend/npu/utils.py
@@ -92,14 +92,6 @@ def init_npu_backend():
     assert _is_npu, "NPU backend initialization called on non-NPU device."
 
     import sgl_kernel_npu  # noqa: F401
-
-    try:
-        import custom_ops  # noqa: F401
-    except ImportError:
-        logger.warning(
-            f"custom_ops not found, dsv3.2 requires this package, which includes the npu_lightning_indexer and npu_sparse_flash_attention operators."
-        )
-
     import torch_npu
     from torch_npu.contrib import transfer_to_npu  # noqa: F401
 
diff --git a/python/sglang/srt/layers/attention/nsa/nsa_indexer.py b/python/sglang/srt/layers/attention/nsa/nsa_indexer.py
index d17523b41..7d7b26d62 100644
--- a/python/sglang/srt/layers/attention/nsa/nsa_indexer.py
+++ b/python/sglang/srt/layers/attention/nsa/nsa_indexer.py
@@ -23,8 +23,7 @@ if _is_cuda:
     except ImportError as e:
         deep_gemm = e
 
-if _is_npu:
-    import custom_ops  # noqa: F401
+if is_npu():
     import torch_npu
     from sglang.srt.hardware_backend.npu.utils import get_indexer_weight_stream
 
@@ -1277,6 +1276,7 @@ class Indexer(MultiPlatformOp):
                 actual_seq_lengths_kv,
                 block_table,
             )
+            return topk_indices
         else:
             block_table = (
                 block_table[: actual_seq_lengths_q.size()[0]]
@@ -1284,7 +1284,7 @@ class Indexer(MultiPlatformOp):
                 else block_table
             )
 
-            topk_indices = torch.ops.custom.npu_lightning_indexer(
+            topk_indices = torch_npu.npu_lightning_indexer(
                 query=q.view(-1, self.n_heads, self.head_dim),
                 key=past_key_states,
                 weights=weights,
@@ -1298,8 +1298,7 @@ class Indexer(MultiPlatformOp):
                 sparse_count=self.index_topk,
                 sparse_mode=3,
             )
-
-        return topk_indices
+            return topk_indices[0]
 
     def do_npu_cp_balance_indexer(
         self,
@@ -1322,7 +1321,7 @@ class Indexer(MultiPlatformOp):
         actual_seq_lengths_q_prev, actual_seq_lengths_q_next = actual_seq_lengths_q
         actual_seq_lengths_kv_prev, actual_seq_lengths_kv_next = actual_seq_lengths_kv
 
-        topk_indices_prev = torch.ops.custom.npu_lightning_indexer(
+        topk_indices_prev = torch_npu.npu_lightning_indexer(
             query=q_prev,
             key=past_key_states,
             weights=weights_prev,
@@ -1338,7 +1337,7 @@ class Indexer(MultiPlatformOp):
             sparse_count=self.index_topk,
             sparse_mode=3,
         )
-        topk_indices_next = torch.ops.custom.npu_lightning_indexer(
+        topk_indices_next = torch_npu.npu_lightning_indexer(
             query=q_next,
             key=past_key_states,
             weights=weights_next,
@@ -1354,4 +1353,4 @@ class Indexer(MultiPlatformOp):
             sparse_count=self.index_topk,
             sparse_mode=3,
         )
-        return topk_indices_prev, topk_indices_next
+        return topk_indices_prev[0], topk_indices_next[0]
diff --git a/scripts/ci/npu/npu_ci_install_dependency.sh b/scripts/ci/npu/npu_ci_install_dependency.sh
index a4ce9bda9..a138e4aa1 100755
--- a/scripts/ci/npu/npu_ci_install_dependency.sh
+++ b/scripts/ci/npu/npu_ci_install_dependency.sh
@@ -7,6 +7,7 @@ DEVICE_TYPE=$1
 
 # Install the required dependencies in CI.
 apt update -y && apt install -y \
+    unzip \
     build-essential \
     cmake \
     wget \
@@ -23,46 +24,34 @@ apt update -y && apt install -y \
 update-ca-certificates
 ${PIP_INSTALL} --upgrade pip
 # Pin wheel to 0.45.1, REF: https://github.com/pypa/wheel/issues/662
-${PIP_INSTALL} wheel==0.45.1 pybind11
+${PIP_INSTALL} wheel==0.45.1 pybind11 pyyaml decorator scipy attrs psutil
 
 
 ### Install MemFabric
-${PIP_INSTALL} memfabric-hybrid==1.0.0
+${PIP_INSTALL} memfabric-hybrid==1.0.5
 
 
 ### Install PyTorch and PTA
 PYTORCH_VERSION="2.8.0"
 TORCHVISION_VERSION="0.23.0"
 ${PIP_INSTALL} torch==${PYTORCH_VERSION} torchvision==${TORCHVISION_VERSION} --index-url https://download.pytorch.org/whl/cpu
-
-PTA_URL="https://sglang-ascend.obs.cn-east-3.myhuaweicloud.com/sglang/torch_npu/torch_npu-2.8.0.post2.dev20251113-cp311-cp311-manylinux_2_28_aarch64.whl"
+PTA_URL="https://gitcode.com/Ascend/pytorch/releases/download/v7.3.0-pytorch2.8.0/torch_npu-2.8.0.post2-cp311-cp311-manylinux_2_28_aarch64.whl"
 ${PIP_INSTALL} ${PTA_URL}
 
 
 ### Install Triton-Ascend
-TRITON_ASCEND_URL="https://sglang-ascend.obs.cn-east-3.myhuaweicloud.com/sglang/triton_ascend/triton_ascend-3.2.0.dev2025112116-cp311-cp311-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl"
-${PIP_INSTALL} ${TRITON_ASCEND_URL}
-
-
-### Install BiSheng
-BISHENG_NAME="Ascend-BiSheng-toolkit_aarch64_20251121.run"
-BISHENG_URL="https://sglang-ascend.obs.cn-east-3.myhuaweicloud.com/sglang/triton_ascend/${BISHENG_NAME}"
-wget -O "${BISHENG_NAME}" "${BISHENG_URL}" && chmod a+x "${BISHENG_NAME}" && "./${BISHENG_NAME}" --install && rm "${BISHENG_NAME}"
+${PIP_INSTALL} triton-ascend
 
 
 ### Install sgl-kernel-npu
-SGL_KERNEL_NPU_TAG="2025.12.31"
-git clone --depth 1 https://github.com/sgl-project/sgl-kernel-npu.git --branch ${SGL_KERNEL_NPU_TAG}
-(cd sgl-kernel-npu && bash ./build.sh && ${PIP_INSTALL} output/deep_ep*.whl output/sgl_kernel_npu*.whl && cd "$(python3 -m pip show deep-ep | grep -E '^Location:' | awk '{print $2}')" && ln -s deep_ep/deep_ep_cpp*.so)
+SGLANG_KERNEL_NPU_TAG="2026.01.21"
+mkdir sgl-kernel-npu
+(cd sgl-kernel-npu && wget https://github.com/sgl-project/sgl-kernel-npu/releases/download/${SGLANG_KERNEL_NPU_TAG}/sgl-kernel-npu_${SGLANG_KERNEL_NPU_TAG}_8.5.0_${DEVICE_TYPE}.zip \
+&& unzip sgl-kernel-npu_${SGLANG_KERNEL_NPU_TAG}_8.5.0_${DEVICE_TYPE}.zip \
+&& ${PIP_INSTALL} output/deep_ep*.whl output/sgl_kernel_npu*.whl \
+&& (cd "$(python3 -m pip show deep-ep | grep -E '^Location:' | awk '{print $2}')" && ln -s deep_ep/deep_ep_cpp*.so))
 
 
-### Install CustomOps (TODO: to be removed once merged into sgl-kernel-npu)
-wget https://sglang-ascend.obs.cn-east-3.myhuaweicloud.com/ops/CANN-custom_ops-8.2.0.0-$DEVICE_TYPE-linux.aarch64.run
-chmod a+x ./CANN-custom_ops-8.2.0.0-$DEVICE_TYPE-linux.aarch64.run
-./CANN-custom_ops-8.2.0.0-$DEVICE_TYPE-linux.aarch64.run --quiet --install-path=/usr/local/Ascend/ascend-toolkit/latest/opp
-wget https://sglang-ascend.obs.cn-east-3.myhuaweicloud.com/ops/custom_ops-1.0.$DEVICE_TYPE-cp311-cp311-linux_aarch64.whl
-pip install ./custom_ops-1.0.$DEVICE_TYPE-cp311-cp311-linux_aarch64.whl
-
 ### Install SGLang
 rm -rf python/pyproject.toml && mv python/pyproject_npu.toml python/pyproject.toml
 ${PIP_INSTALL} -v -e "python[dev_npu]"