diff --git a/tests/examples/online_serving/test_qwen3_omni.py b/tests/examples/online_serving/test_qwen3_omni.py
index 2fbce9dbb..23e2ba184 100644
--- a/tests/examples/online_serving/test_qwen3_omni.py
+++ b/tests/examples/online_serving/test_qwen3_omni.py
@@ -181,9 +181,7 @@ def test_modality_control_003(omni_server) -> None:
 
     # Verify text output same as audio output
     wav_path = extract_last_audio_saved_path(result)
-    # Whisper-small hallucinates a repeated-letter tail on this fixture's WAV.
-    # Use stronger ASR while retaining the keyword and similarity assertions.
-    audio_content = convert_audio_file_to_text(output_path=f"./{wav_path}", model_size="large-v3")
+    audio_content = convert_audio_file_to_text(output_path=f"./{wav_path}")
     print(f"text content is: {text_content}")
     assert "cherry blossom" in audio_content, "The output does not contain any of the keywords."
     print(f"audio content is: {audio_content}")
diff --git a/tests/helpers/tests/test_assertions.py b/tests/helpers/tests/test_assertions.py
index 864f723a4..e2060152f 100644
--- a/tests/helpers/tests/test_assertions.py
+++ b/tests/helpers/tests/test_assertions.py
@@ -3,7 +3,6 @@
 
 import io
 import wave
-from contextlib import nullcontext
 from types import SimpleNamespace
 
 import numpy as np
@@ -20,36 +19,6 @@ from tests.helpers.client import OmniResponse
 pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
 
 
-@pytest.mark.parametrize("result", ["matching", "wrong_content", "missing_keyword"])
-def test_qwen3_omni_documentation_strong_asr_preserves_content_gates(monkeypatch, result):
-    from tests.examples.online_serving import test_qwen3_omni as documentation
-
-    expected = "cherry blossom trees fill the foreground while a tower stands against the blue sky."
-    transcripts = {
-        "matching": expected,
-        "wrong_content": "cherry blossom is unrelated to this recording about baking bread in a kitchen.",
-        "missing_keyword": expected.replace("cherry blossom", "plum flowers"),
-    }
-    monkeypatch.setattr(
-        documentation,
-        "run_cmd",
-        lambda _command: f"Chat completion output from text: {expected}\nAudio saved to retained.wav\n",
-    )
-
-    def transcribe(output_path, model_size="small", language=None):
-        assert output_path == "./retained.wav"
-        return transcripts[result] if model_size == "large-v3" else "f" * 700
-
-    monkeypatch.setattr(documentation, "convert_audio_file_to_text", transcribe)
-    error = {
-        "wrong_content": "The audio content is not same as the text",
-        "missing_keyword": "The output does not contain any of the keywords",
-    }.get(result)
-    expectation = pytest.raises(AssertionError, match=error) if error else nullcontext()
-    with expectation:
-        documentation.test_modality_control_003(SimpleNamespace(model="Qwen/Qwen3-Omni-30B-A3B-Instruct"))
-
-
 def test_short_transcript_repeat_passes_containment_fallback():
     _assert_transcript_matches(
         " How... how are you?",