diff --git a/python/sglang/test/test_utils.py b/python/sglang/test/test_utils.py
index 41e371efe..667a1dedd 100644
--- a/python/sglang/test/test_utils.py
+++ b/python/sglang/test/test_utils.py
@@ -1786,17 +1786,19 @@ def check_evaluation_test_results(
     model_count=None,
 ):
     """
-    results: list of tuple of (model_path, accuracy, latency)
+    results: list of tuple of (model_path, accuracy, latency) or (model_path, accuracy, latency, error)
     """
     failed_models = []
     if model_latency_thresholds is not None:
-        summary = " | model | status | score | score_threshold | latency | latency_threshold | \n"
-        summary += "| ----- | ------ | ----- | --------------- | ------- | ----------------- | \n"
+        summary = " | model | status | score | score_threshold | latency | latency_threshold | error | \n"
+        summary += "| ----- | ------ | ----- | --------------- | ------- | ----------------- | ----- | \n"
     else:
-        summary = " | model | status | score | score_threshold | \n"
-        summary += "| ----- | ------ | ----- | --------------- | \n"
+        summary = " | model | status | score | score_threshold | error | \n"
+        summary += "| ----- | ------ | ----- | --------------- | ----- | \n"
 
-    results_dict = {res[0]: (res[1], res[2]) for res in results}
+    results_dict = {
+        res[0]: (res[1], res[2], res[3] if len(res) == 4 else None) for res in results
+    }
 
     for model, accuracy_threshold in sorted(model_accuracy_thresholds.items()):
         latency_threshold = (
@@ -1805,8 +1807,15 @@ def check_evaluation_test_results(
             else 1e9
         )
 
-        if model in results_dict:
-            accuracy, latency = results_dict[model]
+        # check for error here
+        error = (
+            results_dict.get(model, (None, None, None))[2]
+            if model in results_dict
+            else None
+        )
+
+        if model in results_dict and error is None:
+            accuracy, latency, _ = results_dict[model]
             is_success = accuracy >= accuracy_threshold and latency <= latency_threshold
             status_emoji = "✅" if is_success else "❌"
 
@@ -1823,18 +1832,17 @@ def check_evaluation_test_results(
                     )
 
             if model_latency_thresholds is not None:
-                line = f"| {model} | {status_emoji} | {accuracy} | {accuracy_threshold} | {latency} | {latency_threshold}\n"
+                line = f"| {model} | {status_emoji} | {accuracy} | {accuracy_threshold} | {latency} | {latency_threshold} | - |\n"
             else:
-                line = (
-                    f"| {model} | {status_emoji} | {accuracy} | {accuracy_threshold}\n"
-                )
+                line = f"| {model} | {status_emoji} | {accuracy} | {accuracy_threshold} | - |\n"
         else:
             status_emoji = "❌"
+            error_display = error if error else "Model not evaluated"
             failed_models.append(f"Model failed to launch or be evaluated: {model}")
             if model_latency_thresholds is not None:
-                line = f"| {model} | {status_emoji} | N/A | {accuracy_threshold} | N/A | {latency_threshold}\n"
+                line = f"| {model} | {status_emoji} | N/A | {accuracy_threshold} | N/A | {latency_threshold} | {error_display} |\n"
             else:
-                line = f"| {model} | {status_emoji} | N/A | {accuracy_threshold}\n"
+                line = f"| {model} | {status_emoji} | N/A | {accuracy_threshold} | {error_display} |\n"
 
         summary += line
 
diff --git a/test/nightly/test_text_models_gsm8k_eval.py b/test/nightly/test_text_models_gsm8k_eval.py
index 8cd62e604..c0bf11cea 100644
--- a/test/nightly/test_text_models_gsm8k_eval.py
+++ b/test/nightly/test_text_models_gsm8k_eval.py
@@ -69,6 +69,7 @@ class TestNightlyGsm8KEval(unittest.TestCase):
         for model_setup in self.models:
             with self.subTest(model=model_setup.model_path):
                 other_args = list(model_setup.extra_args)
+                error_message = None
 
                 if model_setup.model_path == "meta-llama/Llama-3.1-70B-Instruct":
                     other_args.extend(["--mem-fraction-static", "0.9"])
@@ -99,8 +100,18 @@ class TestNightlyGsm8KEval(unittest.TestCase):
                     )
                     is_first = False
 
-                    # 0.0 for empty latency
-                    all_results.append((model_setup.model_path, metrics["score"], 0.0))
+                    # 0.0 for empty latency, None for no error
+                    all_results.append(
+                        (model_setup.model_path, metrics["score"], 0.0, error_message)
+                    )
+                except Exception as e:
+                    # Capture error message for the summary table
+                    error_message = str(e)
+                    # Still append result with error info (use None for N/A metrics to match else clause)
+                    all_results.append(
+                        (model_setup.model_path, None, None, error_message)
+                    )
+                    print(f"Error evaluating {model_setup.model_path}: {error_message}")
                 finally:
                     kill_process_tree(process.pid)
 
diff --git a/test/nightly/test_vlms_mmmu_eval.py b/test/nightly/test_vlms_mmmu_eval.py
index 4b9a19eb8..7f9b5aae4 100644
--- a/test/nightly/test_vlms_mmmu_eval.py
+++ b/test/nightly/test_vlms_mmmu_eval.py
@@ -67,6 +67,7 @@ class TestNightlyVLMMmmuEval(unittest.TestCase):
 
         for model in self.models:
             model_path = model.model_path
+            error_message = None
             with self.subTest(model=model_path):
                 process = popen_launch_server(
                     model=model_path,
@@ -98,8 +99,19 @@ class TestNightlyVLMMmmuEval(unittest.TestCase):
                     is_first = False
 
                     all_results.append(
-                        (model_path, metrics["score"], metrics["latency"])
+                        (
+                            model_path,
+                            metrics["score"],
+                            metrics["latency"],
+                            error_message,
+                        )
                     )
+                except Exception as e:
+                    # Capture error message for the summary table
+                    error_message = str(e)
+                    # Still append result with error info (use None for N/A metrics to match else clause)
+                    all_results.append((model_path, None, None, error_message))
+                    print(f"Error evaluating {model_path}: {error_message}")
                 finally:
                     kill_process_tree(process.pid)