diff --git a/tests/test_silero_vad_vendored.py b/tests/test_silero_vad_vendored.py
index e490121c2..7015c9a81 100644
--- a/tests/test_silero_vad_vendored.py
+++ b/tests/test_silero_vad_vendored.py
@@ -6,8 +6,6 @@
 import subprocess
 import sys
 
-import pytest
-
 PROBE = """
 import sys
 import solstone.observe.vad  # noqa: F401
@@ -27,31 +25,3 @@ def test_vendored_module_does_not_pull_faster_whisper():
     )
 
     assert "ok" in result.stdout
-
-
-def test_vendored_get_speech_timestamps_matches_upstream():
-    """Vendored VAD should return the same speech segment bounds as upstream."""
-    upstream_vad = pytest.importorskip("faster_whisper.vad")
-
-    from solstone.observe import _silero_vad as vendored_vad
-    from solstone.observe.utils import SAMPLE_RATE, load_audio
-    from solstone.think.install_models import _fixture_audio_path
-
-    audio = load_audio(_fixture_audio_path())
-    vendored_segments = vendored_vad.get_speech_timestamps(
-        audio,
-        vendored_vad.VadOptions(),
-        sampling_rate=SAMPLE_RATE,
-    )
-    upstream_segments = upstream_vad.get_speech_timestamps(
-        audio,
-        upstream_vad.VadOptions(),
-        sampling_rate=SAMPLE_RATE,
-    )
-
-    assert isinstance(vendored_segments, list)
-    assert isinstance(upstream_segments, list)
-    assert len(vendored_segments) == len(upstream_segments)
-    for vendored, upstream in zip(vendored_segments, upstream_segments):
-        assert vendored["start"] == upstream["start"]
-        assert vendored["end"] == upstream["end"]