diff --git a/.github/workflows/release.yml b/.github/workflows/release.yml
index fb2b578d..95956a0c 100644
--- a/.github/workflows/release.yml
+++ b/.github/workflows/release.yml
@@ -66,6 +66,12 @@ jobs:
         run: |
           pip install -r backend/requirements-mlx.txt
 
+      - name: Install PyTorch with CUDA (Windows only)
+        if: matrix.platform == 'windows-latest'
+        run: |
+          pip install torch --index-url https://download.pytorch.org/whl/cu121 --force-reinstall --no-deps
+          pip install torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
+
       - name: Build Python server (Linux/macOS)
         if: matrix.platform != 'windows-latest'
         run: |
diff --git a/backend/main.py b/backend/main.py
index 83a44bfe..7ac2ed93 100644
--- a/backend/main.py
+++ b/backend/main.py
@@ -1189,10 +1189,11 @@ async def get_model_status():
     if backend_type == "mlx":
         tts_1_7b_id = "mlx-community/Qwen3-TTS-12Hz-1.7B-Base-bf16"
         tts_0_6b_id = "mlx-community/Qwen3-TTS-12Hz-1.7B-Base-bf16"  # Fallback to 1.7B
-        whisper_base_id = "mlx-community/whisper-base"
-        whisper_small_id = "mlx-community/whisper-small"
-        whisper_medium_id = "mlx-community/whisper-medium"
-        whisper_large_id = "mlx-community/whisper-large"
+        # MLX backend uses openai/whisper-* models, not mlx-community
+        whisper_base_id = "openai/whisper-base"
+        whisper_small_id = "openai/whisper-small"
+        whisper_medium_id = "openai/whisper-medium"
+        whisper_large_id = "openai/whisper-large"
     else:
         tts_1_7b_id = "Qwen/Qwen3-TTS-12Hz-1.7B-Base"
         tts_0_6b_id = "Qwen/Qwen3-TTS-12Hz-0.6B-Base"