diff --git a/.github/workflows/accuracy_test.yaml b/.github/workflows/accuracy_test.yaml
index 6d0faf836a..219ad872f2 100644
--- a/.github/workflows/accuracy_test.yaml
+++ b/.github/workflows/accuracy_test.yaml
@@ -57,6 +57,8 @@ jobs:
             model_name: Qwen3-VL-30B-A3B-Instruct
           - runner: a2-2
             model_name: DeepSeek-V2-Lite
+          - runner: a2-2
+            model_name: ERNIE-4.5-21B-A3B-PT
       fail-fast: false
     # test will be triggered when tag 'accuracy-test' & 'ready-for-test'
     if:  >-
diff --git a/tests/e2e/models/configs/ERNIE-4.5-21B-A3B-PT.yaml b/tests/e2e/models/configs/ERNIE-4.5-21B-A3B-PT.yaml
new file mode 100644
index 0000000000..a8c123a432
--- /dev/null
+++ b/tests/e2e/models/configs/ERNIE-4.5-21B-A3B-PT.yaml
@@ -0,0 +1,14 @@
+model_name: "PaddlePaddle/ERNIE-4.5-21B-A3B-PT"
+tasks:
+- name: "gsm8k"
+  metrics:
+  - name: "exact_match,flexible-extract"
+    value: 0.72
+- name: "ceval-valid"
+  metrics:
+  - name: "acc,none"
+    value: 0.85
+num_fewshot: 5
+tensor_parallel_size: 2
+batch_size: 16
+gpu_memory_utilization: 0.6
diff --git a/tests/e2e/models/configs/accuracy.txt b/tests/e2e/models/configs/accuracy.txt
index 3bdcfd8a04..cae04631b0 100644
--- a/tests/e2e/models/configs/accuracy.txt
+++ b/tests/e2e/models/configs/accuracy.txt
@@ -6,3 +6,4 @@ Qwen2-7B.yaml
 Qwen2-VL-7B-Instruct.yaml
 Qwen2-Audio-7B-Instruct.yaml
 Qwen3-VL-30B-A3B-Instruct.yaml
+ERNIE-4.5-21B-A3B-PT.yaml