diff --git a/src/groq/resources/audio/transcriptions.py b/src/groq/resources/audio/transcriptions.py
index 98b70d67..74e41896 100644
--- a/src/groq/resources/audio/transcriptions.py
+++ b/src/groq/resources/audio/transcriptions.py
@@ -25,6 +25,14 @@
 __all__ = ["Transcriptions", "AsyncTranscriptions"]
 
 
+def _get_response_format_type(
+    response_format: Literal["json", "text", "verbose_json"] | Omit,
+) -> type[Transcription] | type[str]:
+    # `text` responses are served as text/plain, so casting to a model makes the parser
+    # fall back to json.loads() and return whatever type the transcript happens to parse as.
+    return str if response_format == "text" else Transcription
+
+
 class Transcriptions(SyncAPIResource):
     @cached_property
     def with_raw_response(self) -> TranscriptionsWithRawResponse:
@@ -167,7 +175,7 @@ def create(
         extra_query: Query | None = None,
         extra_body: Body | None = None,
         timeout: float | httpx.Timeout | None | NotGiven = not_given,
-    ) -> Transcription:
+    ) -> Transcription | str:
         """
         Transcribes audio into the input language.
 
@@ -239,7 +247,7 @@ def create(
             options=make_request_options(
                 extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout
             ),
-            cast_to=Transcription,
+            cast_to=_get_response_format_type(response_format),
         )
 
 
@@ -385,7 +393,7 @@ async def create(
         extra_query: Query | None = None,
         extra_body: Body | None = None,
         timeout: float | httpx.Timeout | None | NotGiven = not_given,
-    ) -> Transcription:
+    ) -> Transcription | str:
         """
         Transcribes audio into the input language.
 
@@ -457,7 +465,7 @@ async def create(
             options=make_request_options(
                 extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout
             ),
-            cast_to=Transcription,
+            cast_to=_get_response_format_type(response_format),
         )
 
 
diff --git a/src/groq/resources/audio/translations.py b/src/groq/resources/audio/translations.py
index 1f0483c1..91ef9183 100644
--- a/src/groq/resources/audio/translations.py
+++ b/src/groq/resources/audio/translations.py
@@ -25,6 +25,14 @@
 __all__ = ["Translations", "AsyncTranslations"]
 
 
+def _get_response_format_type(
+    response_format: Literal["json", "text", "verbose_json"] | Omit,
+) -> type[Translation] | type[str]:
+    # `text` responses are served as text/plain, so casting to a model makes the parser
+    # fall back to json.loads() and return whatever type the transcript happens to parse as.
+    return str if response_format == "text" else Translation
+
+
 class Translations(SyncAPIResource):
     @cached_property
     def with_raw_response(self) -> TranslationsWithRawResponse:
@@ -60,7 +68,7 @@ def create(
         extra_query: Query | None = None,
         extra_body: Body | None = None,
         timeout: float | httpx.Timeout | None | NotGiven = not_given,
-    ) -> Translation:
+    ) -> Translation | str:
         """Translates audio into English.
 
         Args:
@@ -119,7 +127,7 @@ def create(
             options=make_request_options(
                 extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout
             ),
-            cast_to=Translation,
+            cast_to=_get_response_format_type(response_format),
         )
 
 
@@ -158,7 +166,7 @@ async def create(
         extra_query: Query | None = None,
         extra_body: Body | None = None,
         timeout: float | httpx.Timeout | None | NotGiven = not_given,
-    ) -> Translation:
+    ) -> Translation | str:
         """Translates audio into English.
 
         Args:
@@ -217,7 +225,7 @@ async def create(
             options=make_request_options(
                 extra_headers=extra_headers, extra_query=extra_query, extra_body=extra_body, timeout=timeout
             ),
-            cast_to=Translation,
+            cast_to=_get_response_format_type(response_format),
         )
 
 
diff --git a/tests/test_audio_response_format.py b/tests/test_audio_response_format.py
new file mode 100644
index 00000000..1cfe3a29
--- /dev/null
+++ b/tests/test_audio_response_format.py
@@ -0,0 +1,91 @@
+from __future__ import annotations
+
+import os
+
+import httpx
+import pytest
+from respx import MockRouter
+
+from groq import Groq, AsyncGroq
+from groq.types.audio import Translation, Transcription
+
+base_url = os.environ.get("TEST_API_BASE_URL", "http://127.0.0.1:4010")
+
+# A transcript is arbitrary user speech, so it may happen to be valid JSON.
+TEXT_BODIES = ["Hello there.", "42", "true", "null", "3.5", "[1, 2]", '{"text": "hi"}']
+
+
+def _text_response(body: str) -> httpx.Response:
+    return httpx.Response(200, headers={"Content-Type": "text/plain; charset=utf-8"}, content=body)
+
+
+class TestTranscriptions:
+    @pytest.mark.parametrize("body", TEXT_BODIES)
+    @pytest.mark.respx(base_url=base_url)
+    def test_response_format_text_returns_str(self, body: str, respx_mock: MockRouter, client: Groq) -> None:
+        respx_mock.post("/openai/v1/audio/transcriptions").mock(return_value=_text_response(body))
+
+        transcription = client.audio.transcriptions.create(
+            model="whisper-large-v3", file=("audio.wav", b"RIFF"), response_format="text"
+        )
+        assert transcription == body
+
+    @pytest.mark.respx(base_url=base_url)
+    def test_response_format_json_returns_model(self, respx_mock: MockRouter, client: Groq) -> None:
+        respx_mock.post("/openai/v1/audio/transcriptions").mock(
+            return_value=httpx.Response(200, json={"text": "Hello there."})
+        )
+
+        transcription = client.audio.transcriptions.create(
+            model="whisper-large-v3", file=("audio.wav", b"RIFF"), response_format="json"
+        )
+        assert isinstance(transcription, Transcription)
+        assert transcription.text == "Hello there."
+
+    @pytest.mark.parametrize("body", TEXT_BODIES)
+    @pytest.mark.respx(base_url=base_url)
+    async def test_async_response_format_text_returns_str(
+        self, body: str, respx_mock: MockRouter, async_client: AsyncGroq
+    ) -> None:
+        respx_mock.post("/openai/v1/audio/transcriptions").mock(return_value=_text_response(body))
+
+        transcription = await async_client.audio.transcriptions.create(
+            model="whisper-large-v3", file=("audio.wav", b"RIFF"), response_format="text"
+        )
+        assert transcription == body
+
+
+class TestTranslations:
+    @pytest.mark.parametrize("body", TEXT_BODIES)
+    @pytest.mark.respx(base_url=base_url)
+    def test_response_format_text_returns_str(self, body: str, respx_mock: MockRouter, client: Groq) -> None:
+        respx_mock.post("/openai/v1/audio/translations").mock(return_value=_text_response(body))
+
+        translation = client.audio.translations.create(
+            model="whisper-large-v3", file=("audio.wav", b"RIFF"), response_format="text"
+        )
+        assert translation == body
+
+    @pytest.mark.respx(base_url=base_url)
+    def test_response_format_json_returns_model(self, respx_mock: MockRouter, client: Groq) -> None:
+        respx_mock.post("/openai/v1/audio/translations").mock(
+            return_value=httpx.Response(200, json={"text": "Hello there."})
+        )
+
+        translation = client.audio.translations.create(
+            model="whisper-large-v3", file=("audio.wav", b"RIFF"), response_format="json"
+        )
+        assert isinstance(translation, Translation)
+        assert translation.text == "Hello there."
+
+    @pytest.mark.parametrize("body", TEXT_BODIES)
+    @pytest.mark.respx(base_url=base_url)
+    async def test_async_response_format_text_returns_str(
+        self, body: str, respx_mock: MockRouter, async_client: AsyncGroq
+    ) -> None:
+        respx_mock.post("/openai/v1/audio/translations").mock(return_value=_text_response(body))
+
+        translation = await async_client.audio.translations.create(
+            model="whisper-large-v3", file=("audio.wav", b"RIFF"), response_format="text"
+        )
+        assert translation == body