diff --git a/tests/models/multimodal/processing/test_tensor_schema.py b/tests/models/multimodal/processing/test_tensor_schema.py
index 1a11fa3d2b82..615564f70ea3 100644
--- a/tests/models/multimodal/processing/test_tensor_schema.py
+++ b/tests/models/multimodal/processing/test_tensor_schema.py
@@ -41,9 +41,6 @@ ARCH_NEEDS_EXTRAS = [
 ]
 REPO_ID_TO_SKIP = {
     "nm-testing/pixtral-12b-FP8-dynamic": "duplicated test",
-    # FIXME(Isotr0py): enable GPT-OSS based InternVL3.5 model
-    # after support PP for GPT-OSS
-    "OpenGVLab/InternVL3_5-GPT-OSS-20B-A4B-Preview": "Broken model",
 }
 
 ImageInput = list[Image.Image]
diff --git a/vllm/multimodal/utils.py b/vllm/multimodal/utils.py
index 794e24c2c748..e09c97de576e 100644
--- a/vllm/multimodal/utils.py
+++ b/vllm/multimodal/utils.py
@@ -378,10 +378,7 @@ def group_mm_inputs_by_modality(
         elif len(mm_input) == 1:
             return next(iter(mm_input.keys()))
 
-        # FIXME(Isotr0py): Modality of mm_input from legacy pipeline is empty,
-        # this is used to make InternVL with legacy pipeline still work with v1.
-        else:
-            return ""
+        raise AssertionError("This line should be unreachable.")
 
     return [
         list(group) for _, group in groupby(mm_inputs, key=modality_group_func)