diff --git a/scenedetect/backends/opencv.py b/scenedetect/backends/opencv.py
index 9e2ab094..5abe78c6 100644
--- a/scenedetect/backends/opencv.py
+++ b/scenedetect/backends/opencv.py
@@ -238,6 +238,8 @@ def timecode(self) -> Timecode:
         # *NOTE*: Although OpenCV has `CAP_PROP_PTS`, it doesn't seem to be reliable. For now, we
         # use `CAP_PROP_POS_MSEC` instead, converting to microseconds for sufficient precision to
         # avoid frame-boundary rounding errors at common framerates like 24000/1001.
+        if self._eof_timecode is not None:
+            return self._eof_timecode
         ms = self._cap.get(cv2.CAP_PROP_POS_MSEC)
         time_base = Fraction(1, 1000000)
         return Timecode(pts=round(ms * 1000), time_base=time_base)
@@ -272,6 +274,7 @@ def seek(self, target: TimecodeLike):
             raise ValueError("Target seek position cannot be negative!")
         target_secs = (self.base_timecode + target).seconds
         self._has_grabbed = False
+        self._eof_timecode = None
         if target_secs > 0:
             # Seek one frame before target so the next read() returns the frame at target.
             one_frame_ms = 1000.0 / float(self._frame_rate)
@@ -310,6 +313,9 @@ def reset(self):
     def read(self, decode: bool = True) -> np.ndarray | bool:
         if not self._cap.isOpened():
             return False
+        # At the end of some VFR streams OpenCV reports an earlier CAP_PROP_POS_MSEC once grab()
+        # fails, so keep `position` on the last frame read like the other backends do (#561).
+        last_timecode = self.timecode if self._has_grabbed else None
         has_grabbed = self._cap.grab()
         # If we failed to grab the frame, retry a few times if required.
         if not has_grabbed:
@@ -327,8 +333,10 @@ def read(self, decode: bool = True) -> np.ndarray | bool:
                     logger.warning("Failed to decode some frames, results may be inaccurate.")
         # We didn't manage to grab a frame even after retrying, so just return.
         if not has_grabbed:
+            self._eof_timecode = last_timecode
             return False
         self._has_grabbed = True
+        self._eof_timecode = None
         # Need to make sure we actually grabbed a frame before calling retrieve.
         if decode and self._has_grabbed:
             _, frame = self._cap.retrieve()
@@ -391,6 +399,7 @@ def _open_capture(self, frame_rate: FrameRate | None = None):
         self._cap: cv2.VideoCapture = cap
         self._frame_rate: Fraction = framerate_to_fraction(frame_rate)
         self._has_grabbed = False
+        self._eof_timecode: Timecode | None = None
         cap.set(cv2.CAP_PROP_ORIENTATION_AUTO, 1.0)  # https://github.com/opencv/opencv/issues/26795
 
 
diff --git a/tests/test_vfr.py b/tests/test_vfr.py
index 0a6989e1..ce8589c1 100644
--- a/tests/test_vfr.py
+++ b/tests/test_vfr.py
@@ -95,6 +95,16 @@ def test_vfr_position_monotonic_opencv(test_vfr_video: str, auto_close):
     assert frame_count > 0
 
 
+def test_vfr_position_after_eof_opencv(test_vfr_drop3_video: str, auto_close):
+    """Position should stay on the last frame once read() hits the end of the video (#561)."""
+    video = auto_close(open_video(test_vfr_drop3_video, backend="opencv"))
+    last_position = None
+    while video.read() is not False:
+        last_position = video.position
+    assert last_position is not None
+    assert video.position == last_position
+
+
 @pytest.mark.parametrize("backend", ["pyav", "opencv"])
 def test_vfr_scene_detection(test_vfr_video: str, backend: str, auto_close):
     """Scene detection on VFR video should produce timestamps matching known ground truth.
diff --git a/website/pages/changelog.md b/website/pages/changelog.md
index a1c6fbbe..bce462df 100644
--- a/website/pages/changelog.md
+++ b/website/pages/changelog.md
@@ -6,6 +6,7 @@
 ### Unreleased
 
  - [bugfix] OTIO exports use the full source duration when provided, including from the CLI, instead of declaring a fixed 1,980-frame duration. The available range remains unspecified when the duration is unknown. Scene source ranges are unchanged. [#580](https://github.com/Breakthrough/PySceneDetect/pull/580)
+ - [bugfix] The OpenCV backend keeps `position` on the last frame after reaching the end of a VFR video, so the last scene no longer ends early (or before it starts) [#561](https://github.com/Breakthrough/PySceneDetect/issues/561)
 
 ### PySceneDetect 0.7.2 (October 11, 2026)