fix(anthropic): convert PDF image_url content to documents

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

litellm/litellm_core_utils/prompt_templates/factory.py

@@ -907,9 +907,9 @@ def create_anthropic_image_param(
     image_url_input: str | dict,
     format: str | None = None,
     is_bedrock_invoke: bool = False,
-) -> AnthropicMessagesImageParam:
+) -> AnthropicMessagesImageParam | AnthropicMessagesDocumentParam:
     """
-    Create an AnthropicMessagesImageParam from an image URL input.
+    Create an Anthropic content block from an image URL input.
 
     Supports both URL references (for HTTP/HTTPS URLs) and base64 encoding.
     """
@@ -928,14 +928,7 @@ def create_anthropic_image_param(
         if is_bedrock_invoke or image_url.startswith("http://"):
             base64_url: Final = convert_url_to_base64(url=image_url)
             image_chunk = convert_to_anthropic_image_obj(openai_image_url=base64_url, format=format)
-            return AnthropicMessagesImageParam(
-                type="image",
-                source=AnthropicContentParamSource(
-                    type="base64",
-                    media_type=image_chunk["media_type"],
-                    data=image_chunk["data"],
-                ),
-            )
+            return _anthropic_content_element_factory(image_chunk)
         else:
             # HTTPS URL - pass directly for regular Anthropic
             return AnthropicMessagesImageParam(
@@ -948,14 +941,7 @@ def create_anthropic_image_param(
     else:
         # Convert to base64 for data URIs or other formats
         image_chunk = convert_to_anthropic_image_obj(openai_image_url=image_url, format=format)
-        return AnthropicMessagesImageParam(
-            type="image",
-            source=AnthropicContentParamSource(
-                type="base64",
-                media_type=image_chunk["media_type"],
-                data=image_chunk["data"],
-            ),
-        )
+        return _anthropic_content_element_factory(image_chunk)
 
 
 # The following XML functions will be deprecated once JSON schema support is available on Bedrock and Vertex
@@ -1795,7 +1781,12 @@ def convert_to_anthropic_tool_result(
                         anthropic_content_element=_anthropic_image_param,
                         original_content_element=content,
                     )
-                    anthropic_content_list.append(cast(AnthropicMessagesImageParam, _anthropic_image_param))
+                    anthropic_content_list.append(
+                        cast(
+                            AnthropicMessagesImageParam | AnthropicMessagesDocumentParam,
+                            _anthropic_image_param,
+                        )
+                    )
             elif content["type"] == "tool_reference":
                 anthropic_content_list.append(ToolReference(type="tool_reference", tool_name=content["tool_name"]))
             elif content["type"] == "file":

tests/unit/litellm_core_utils/prompt_templates/test_litellm_core_utils_prompt_templates_factory.py

@@ -867,6 +867,80 @@ def test_anthropic_messages_pt_keeps_signed_thinking_block():
     assert thinking_blocks[0]["thinking"] == "genuine anthropic reasoning"
 
 
+@pytest.mark.parametrize("llm_provider", ("anthropic", "vertex_ai"))
+def test_anthropic_messages_pt_converts_pdf_image_url_data_uri_to_document(llm_provider: str) -> None:
+    pdf_data = "JVBERi0xLjQK"
+    result: Final = anthropic_messages_pt(
+        messages=[
+            {
+                "role": "user",
+                "content": [
+                    {"type": "text", "text": "summarize"},
+                    {"type": "image_url", "image_url": {"url": f"data:application/pdf;base64,{pdf_data}"}},
+                ],
+            }
+        ],
+        model="claude-opus-5-5",
+        llm_provider=llm_provider,
+    )
+
+    assert result[0]["content"][1] == {
+        "type": "document",
+        "source": {"type": "base64", "media_type": "application/pdf", "data": pdf_data},
+    }
+
+
+@pytest.mark.parametrize("llm_provider", ("anthropic", "vertex_ai"))
+def test_anthropic_messages_pt_uses_pdf_format_for_generic_mime_data_uri(llm_provider: str) -> None:
+    pdf_data = "JVBERi0xLjQK"
+    result: Final = anthropic_messages_pt(
+        messages=[
+            {
+                "role": "user",
+                "content": [
+                    {"type": "text", "text": "summarize"},
+                    {
+                        "type": "image_url",
+                        "image_url": {
+                            "url": f"data:application/octet-stream;base64,{pdf_data}",
+                            "format": "application/pdf",
+                        },
+                    },
+                ],
+            }
+        ],
+        model="claude-opus-5-5",
+        llm_provider=llm_provider,
+    )
+
+    assert result[0]["content"][1] == {
+        "type": "document",
+        "source": {"type": "base64", "media_type": "application/pdf", "data": pdf_data},
+    }
+
+
+def test_anthropic_messages_pt_keeps_png_image_url_as_image() -> None:
+    image_data = "iVBORw0KGgo="
+    result: Final = anthropic_messages_pt(
+        messages=[
+            {
+                "role": "user",
+                "content": [
+                    {"type": "text", "text": "describe"},
+                    {"type": "image_url", "image_url": f"data:image/png;base64,{image_data}"},
+                ],
+            }
+        ],
+        model="claude-opus-5-5",
+        llm_provider="anthropic",
+    )
+
+    assert result[0]["content"][1] == {
+        "type": "image",
+        "source": {"type": "base64", "media_type": "image/png", "data": image_data},
+    }
+
+
 EMPTY_USER_MESSAGES: Final = pytest.mark.parametrize(
     "empty_user_message",
     [{"role": "user", "content": None}, {"role": "user", "content": []}, {"role": "user"}],