diff --git a/test/srt/run_suite.py b/test/srt/run_suite.py
index a7d5b6744b3a..6192d2a67750 100644
--- a/test/srt/run_suite.py
+++ b/test/srt/run_suite.py
@@ -72,12 +72,14 @@
 }
 
 # Add Intel XPU tests
-# NOTE: please sort the test cases alphabetically by the test file name
+# NOTE: Intentionally NOT alphabetical. Lighter benchmarks run first because
+# heavy models (e.g. DeepSeek-OCR ~6GB) can leave XPU device memory unreclaimed,
+# causing OOM for subsequent tests on memory-constrained devices.
 suite_xpu = {
     "per-commit-xpu": [
+        TestFile("xpu/test_intel_xpu_backend.py"),
         TestFile("xpu/test_deepseek_ocr.py"),
         # TestFile("xpu/test_internvl.py"),
-        TestFile("xpu/test_intel_xpu_backend.py"),
     ],
 }
 
diff --git a/test/srt/xpu/test_deepseek_ocr.py b/test/srt/xpu/test_deepseek_ocr.py
index 9d5da10e363a..e5d1c3fccb83 100644
--- a/test/srt/xpu/test_deepseek_ocr.py
+++ b/test/srt/xpu/test_deepseek_ocr.py
@@ -11,6 +11,7 @@
 
 from sglang.srt.utils import kill_process_tree
 from sglang.test.test_utils import (
+    DEFAULT_IMAGE_URL,
     DEFAULT_TIMEOUT_FOR_SERVER_LAUNCH,
     DEFAULT_URL_FOR_TEST,
     CustomTestCase,
@@ -50,7 +51,7 @@ def get_request_json(self, max_new_tokens=32, n=1):
             self.base_url + "/generate",
             json={
                 "text": "<image>\n<|grounding|>Convert the document to pure text.",
-                "image_data": "../../examples/assets/example_image.png",
+                "image_data": DEFAULT_IMAGE_URL,
                 "sampling_params": {
                     "temperature": 0 if n == 1 else 0.5,
                     "max_new_tokens": max_new_tokens,