From 01556557f4e40805866fa166ed1580241ea35d6d Mon Sep 17 00:00:00 2001
From: cohml <62400541+cohml@users.noreply.github.com>
Date: Tue, 9 Jun 2026 22:34:23 -0400
Subject: [PATCH] fix(audio): don't override explicit response_format with
 verbose_json

---
 .../transcriptions/whisper_transformation.py  |  4 +-
 .../test_whisper_transformation.py            | 43 +++++++++++++++++++
 2 files changed, 44 insertions(+), 3 deletions(-)
 create mode 100644 tests/test_litellm/llms/openai/transcriptions/test_whisper_transformation.py

diff --git a/litellm/llms/openai/transcriptions/whisper_transformation.py b/litellm/llms/openai/transcriptions/whisper_transformation.py
index fa507e1bc26..170b90a3bdb 100644
--- a/litellm/llms/openai/transcriptions/whisper_transformation.py
+++ b/litellm/llms/openai/transcriptions/whisper_transformation.py
@@ -107,9 +107,7 @@ class OpenAIWhisperAudioTranscriptionConfig(BaseAudioTranscriptionConfig):
         """
         data = {"model": model, "file": audio_file, **optional_params}
 
-        if "response_format" not in data or (
-            data["response_format"] == "text" or data["response_format"] == "json"
-        ):
+        if "response_format" not in data:
             data["response_format"] = (
                 "verbose_json"  # ensures 'duration' is received - used for cost calculation
             )
diff --git a/tests/test_litellm/llms/openai/transcriptions/test_whisper_transformation.py b/tests/test_litellm/llms/openai/transcriptions/test_whisper_transformation.py
new file mode 100644
index 00000000000..808e29a2d6f
--- /dev/null
+++ b/tests/test_litellm/llms/openai/transcriptions/test_whisper_transformation.py
@@ -0,0 +1,43 @@
+"""
+Tests for OpenAIWhisperAudioTranscriptionConfig.transform_audio_transcription_request.
+"""
+
+import io
+
+from litellm.llms.openai.transcriptions.whisper_transformation import (
+    OpenAIWhisperAudioTranscriptionConfig,
+)
+
+
+class TestWhisperTransformRequestResponseFormat:
+    def _transform(self, optional_params: dict) -> dict:
+        config = OpenAIWhisperAudioTranscriptionConfig()
+        audio_file = io.BytesIO(b"fake audio")
+        audio_file.name = "test.wav"
+        result = config.transform_audio_transcription_request(
+            model="whisper-1",
+            audio_file=audio_file,
+            optional_params=optional_params,
+            litellm_params={},
+        )
+        return result.data
+
+    def test_defaults_to_verbose_json_when_unset(self):
+        """When response_format is not specified, default to verbose_json for cost calculation."""
+        data = self._transform({})
+        assert data["response_format"] == "verbose_json"
+
+    def test_respects_explicit_json(self):
+        """When response_format='json' is set, do not override to verbose_json."""
+        data = self._transform({"response_format": "json"})
+        assert data["response_format"] == "json"
+
+    def test_respects_explicit_text(self):
+        """When response_format='text' is set, do not override to verbose_json."""
+        data = self._transform({"response_format": "text"})
+        assert data["response_format"] == "text"
+
+    def test_preserves_verbose_json_when_set(self):
+        """verbose_json explicitly set by the caller stays as-is."""
+        data = self._transform({"response_format": "verbose_json"})
+        assert data["response_format"] == "verbose_json"