fix(gemini): handle Gemini Files API URIs in _process_gemini_media without mime type resolution

Gemini Files API URIs (https://generativelanguage.googleapis.com/v1beta/files/*)
have no file extension and return 403 for unauthenticated requests, causing
litellm to crash when trying to sniff the mime type. Add an explicit branch
that forwards these URIs as file_data without attempting mime type resolution,
since the Files API already knows the type of the uploaded file.

Fixes #24907

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

litellm/llms/vertex_ai/gemini/transformation.py

@@ -208,6 +208,14 @@ def _process_gemini_media(
             return _apply_gemini_3_metadata(
                 part, model, media_resolution_enum, video_metadata
             )
+        elif "https://generativelanguage.googleapis.com/v1beta/files/" in image_url:
+            # Gemini Files API URI — mime type is already known by the API,
+            # no need to resolve it from the URL (which would cause a 403).
+            file_data = FileDataType(file_uri=image_url)
+            part = {"file_data": file_data}
+            return _apply_gemini_3_metadata(
+                part, model, media_resolution_enum, video_metadata
+            )
         elif (
             "https://" in image_url
             and (image_type := format or _get_image_mime_type_from_url(image_url))

tests/test_litellm/llms/vertex_ai/gemini/test_vertex_ai_gemini_transformation.py

@@ -3,6 +3,7 @@ from litellm.litellm_core_utils.prompt_templates.factory import (
 )
 from litellm.llms.vertex_ai.gemini.transformation import (
     _gemini_convert_messages_with_history,
+    _process_gemini_media,
     _transform_request_body,
     check_if_part_exists_in_parts,
     _get_highest_media_resolution,
@@ -1812,3 +1813,39 @@ def test_multi_turn_function_calling_roles():
                 assert (
                     content["role"] == "user"
                 ), f"Content block {i} with function_response has role='{content['role']}', expected 'user'"
+
+
+def test_process_gemini_media_files_api_uri_no_mime_type_required():
+    """
+    Regression test for https://github.com/BerriAI/litellm/issues/24907.
+
+    Gemini Files API URIs (https://generativelanguage.googleapis.com/v1beta/files/*)
+    must be forwarded as file_data without attempting to resolve a mime type from the
+    URL. The URL has no file extension and the endpoint returns 403 for unauthenticated
+    requests, so any attempt to download or sniff the mime type raises an error.
+    """
+    gemini_file_uri = (
+        "https://generativelanguage.googleapis.com/v1beta/files/37eh7rsw1vfe"
+    )
+
+    part = _process_gemini_media(image_url=gemini_file_uri)
+
+    assert "file_data" in part
+    assert part["file_data"]["file_uri"] == gemini_file_uri
+    # mime_type must NOT be set — the Files API already knows the type
+    assert part["file_data"].get("mime_type") is None
+
+
+def test_process_gemini_media_files_api_uri_with_explicit_format():
+    """
+    When the caller explicitly provides a format/mime type for a Gemini Files API URI,
+    it should still be forwarded as file_data (not attempted as an inline download).
+    """
+    gemini_file_uri = (
+        "https://generativelanguage.googleapis.com/v1beta/files/n1vhxa28lyaw"
+    )
+
+    part = _process_gemini_media(image_url=gemini_file_uri, format="text/plain")
+
+    assert "file_data" in part
+    assert part["file_data"]["file_uri"] == gemini_file_uri