diff --git a/.github/workflows/windows_wheel.yaml b/.github/workflows/windows_wheel.yaml
index 38cc53be9..b594c54dc 100644
--- a/.github/workflows/windows_wheel.yaml
+++ b/.github/workflows/windows_wheel.yaml
@@ -71,8 +71,7 @@ jobs:
         # TODO: FFmpeg 5 on Windows segfaults in avcodec_open2() when passing
         # bad parameters.
         # See https://github.com/pytorch/torchcodec/pull/806
-        # TODO: Support FFmpeg 8 on Windows
-        ffmpeg-version-for-tests: ['4.4.2', '6.1.1', '7.0.1']
+        ffmpeg-version-for-tests: ['4.4.2', '6.1.1', '7.0.1', '8.0']
     needs: build
     steps:
       - uses: actions/download-artifact@v4
@@ -83,7 +82,11 @@ jobs:
         uses: conda-incubator/setup-miniconda@v2
         with:
           auto-update-conda: true
-          miniconda-version: "latest"
+          # Using miniforge instead of miniconda ensures that the default
+          # conda channel is conda-forge instead of main/default. This ensures
+          # ABI consistency between dependencies:
+          # https://conda-forge.org/docs/user/transitioning_from_defaults/
+          miniforge-version: latest
           activate-environment: test
           python-version: ${{ matrix.python-version }}
       - name: Update pip
diff --git a/test/test_encoders.py b/test/test_encoders.py
index f8b5b3519..c5946654d 100644
--- a/test/test_encoders.py
+++ b/test/test_encoders.py
@@ -16,6 +16,7 @@
 from .utils import (
     assert_tensor_close_on_at_least,
     get_ffmpeg_major_version,
+    get_ffmpeg_minor_version,
     in_fbcode,
     IS_WINDOWS,
     NASA_AUDIO_MP3,
@@ -23,6 +24,11 @@
     TestContainerFile,
 )
 
+IS_WINDOWS_WITH_FFMPEG_LE_70 = IS_WINDOWS and (
+    get_ffmpeg_major_version() < 7
+    or (get_ffmpeg_major_version() == 7 and get_ffmpeg_minor_version() == 0)
+)
+
 
 @pytest.fixture
 def with_ffmpeg_debug_logs():
@@ -155,7 +161,11 @@ def test_bad_input_parametrized(self, method, tmp_path):
         avcodec_open2_failed_msg = "avcodec_open2 failed: Invalid argument"
         with pytest.raises(
             RuntimeError,
-            match=avcodec_open2_failed_msg if IS_WINDOWS else "invalid sample rate=10",
+            match=(
+                avcodec_open2_failed_msg
+                if IS_WINDOWS_WITH_FFMPEG_LE_70
+                else "invalid sample rate=10"
+            ),
         ):
             getattr(decoder, method)(**valid_params)
 
@@ -164,14 +174,18 @@ def test_bad_input_parametrized(self, method, tmp_path):
         )
         with pytest.raises(
             RuntimeError,
-            match=avcodec_open2_failed_msg if IS_WINDOWS else "invalid sample rate=10",
+            match=(
+                avcodec_open2_failed_msg
+                if IS_WINDOWS_WITH_FFMPEG_LE_70
+                else "invalid sample rate=10"
+            ),
         ):
             getattr(decoder, method)(sample_rate=10, **valid_params)
         with pytest.raises(
             RuntimeError,
             match=(
                 avcodec_open2_failed_msg
-                if IS_WINDOWS
+                if IS_WINDOWS_WITH_FFMPEG_LE_70
                 else "invalid sample rate=99999999"
             ),
         ):
@@ -192,7 +206,7 @@ def test_bad_input_parametrized(self, method, tmp_path):
         for num_channels in (0, 3):
             match = (
                 avcodec_open2_failed_msg
-                if IS_WINDOWS
+                if IS_WINDOWS_WITH_FFMPEG_LE_70
                 else re.escape(
                     f"Desired number of channels ({num_channels}) is not supported"
                 )
@@ -316,7 +330,7 @@ def test_against_cli(
         else:
             rtol, atol = None, None
 
-        if IS_WINDOWS and format == "mp3":
+        if IS_WINDOWS_WITH_FFMPEG_LE_70 and format == "mp3":
             # We're getting a "Could not open input file" on Windows mp3 files when decoding.
             # TODO: https://github.com/pytorch/torchcodec/issues/837
             return
@@ -370,7 +384,7 @@ def test_against_to_file(
         else:
             raise ValueError(f"Unknown method: {method}")
 
-        if not (IS_WINDOWS and format == "mp3"):
+        if not (IS_WINDOWS_WITH_FFMPEG_LE_70 and format == "mp3"):
             # We're getting a "Could not open input file" on Windows mp3 files when decoding.
             # TODO: https://github.com/pytorch/torchcodec/issues/837
             torch.testing.assert_close(
diff --git a/test/utils.py b/test/utils.py
index 61bf06295..e11411bd2 100644
--- a/test/utils.py
+++ b/test/utils.py
@@ -76,16 +76,27 @@ def make_video_decoder(*args, **kwargs) -> tuple[VideoDecoder, str]:
     return dec, clean_device
 
 
-def get_ffmpeg_major_version():
+def _get_ffmpeg_version_string():
     ffmpeg_version = get_ffmpeg_library_versions()["ffmpeg_version"]
     # When building FFmpeg from source there can be a `n` prefix in the version
     # string.  This is quite brittle as we're using av_version_info(), which has
     # no stable format. See https://github.com/pytorch/torchcodec/issues/100
     if ffmpeg_version.startswith("n"):
         ffmpeg_version = ffmpeg_version.removeprefix("n")
+
+    return ffmpeg_version
+
+
+def get_ffmpeg_major_version():
+    ffmpeg_version = _get_ffmpeg_version_string()
     return int(ffmpeg_version.split(".")[0])
 
 
+def get_ffmpeg_minor_version():
+    ffmpeg_version = _get_ffmpeg_version_string()
+    return int(ffmpeg_version.split(".")[1])
+
+
 def cuda_version_used_for_building_torch() -> Optional[tuple[int, int]]:
     # Return the CUDA version that was used to build PyTorch. That's not always
     # the same as the CUDA version that is currently installed on the running