diff --git a/tests/model_executor/models/minicpmo_4_5/test_minicpmo_4_5_omni_llm_embed.py b/tests/model_executor/models/minicpmo_4_5/test_minicpmo_4_5_omni_llm_embed.py
new file mode 100644
index 000000000..9f53fb7ec
--- /dev/null
+++ b/tests/model_executor/models/minicpmo_4_5/test_minicpmo_4_5_omni_llm_embed.py
@@ -0,0 +1,33 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+
+import pytest
+
+from vllm_omni.model_executor.models.minicpmo_4_5.minicpmo_4_5_omni_llm import (
+    MiniCPMO45OmniLLMForConditionalGeneration,
+)
+
+pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
+
+
+def test_embed_multimodal_delegates_to_get_multimodal_embeddings(monkeypatch):
+    """vLLM V1 encoder profiling calls embed_multimodal; without an override the
+    SupportsMultiModal protocol stub returns None and startup fails downstream."""
+    sentinel = (object(),)
+    seen: dict[str, object] = {}
+
+    def fake_get_multimodal_embeddings(self, **kwargs):
+        seen.update(kwargs)
+        return sentinel
+
+    monkeypatch.setattr(
+        MiniCPMO45OmniLLMForConditionalGeneration,
+        "get_multimodal_embeddings",
+        fake_get_multimodal_embeddings,
+    )
+    model = object.__new__(MiniCPMO45OmniLLMForConditionalGeneration)
+
+    out = model.embed_multimodal(pixel_values="pv", audio_features="af")
+
+    assert out is sentinel
+    assert seen == {"pixel_values": "pv", "audio_features": "af"}
diff --git a/vllm_omni/model_executor/models/minicpmo_4_5/minicpmo_4_5_omni_llm.py b/vllm_omni/model_executor/models/minicpmo_4_5/minicpmo_4_5_omni_llm.py
index d70bc30d0..6d1c5de32 100644
--- a/vllm_omni/model_executor/models/minicpmo_4_5/minicpmo_4_5_omni_llm.py
+++ b/vllm_omni/model_executor/models/minicpmo_4_5/minicpmo_4_5_omni_llm.py
@@ -4580,6 +4580,10 @@ class MiniCPMO45OmniLLMForConditionalGeneration(nn.Module, SupportsMultiModal, S
                 multimodal_embeddings += tuple(audio_embeddings)
         return multimodal_embeddings
 
+    def embed_multimodal(self, **kwargs: object) -> MultiModalEmbeddings:
+        """vLLM V1 encoder profiling calls this; the inherited Protocol stub returns None."""
+        return self.get_multimodal_embeddings(**kwargs)
+
     def get_input_embeddings(
         self,
         input_ids: torch.Tensor,