fix(core): use parsed mime_type for base64 file blocks in openai translator

libs/core/langchain_core/messages/block_translators/openai.py

@@ -94,6 +94,15 @@ def convert_to_openai_data_block(
         if block.get("source_type") == "base64" or "base64" in block:
             # Handle v0 format (Base64CB): {"source_type": "base64", "data": "...", ...}
             # Handle v1 format (IDCB): {"base64": "...", ...}
+            if (
+                api == "chat/completions"
+                and block.get("mime_type") != "application/pdf"
+            ):
+                error_msg = (
+                    "OpenAI Chat Completions only supports application/pdf for "
+                    "file data blocks. Use Responses API for other file MIME types."
+                )
+                raise ValueError(error_msg)
             base64_data = block["data"] if "source_type" in block else block["base64"]
             file = {"file_data": f"data:{block['mime_type']};base64,{base64_data}"}
             if filename := block.get("filename"):
@@ -548,7 +557,7 @@ def _extract_extras(
         filename = block["file"].get("filename")
         return types.create_file_block(
             base64=parsed["data"],
-            mime_type="application/pdf",
+            mime_type=parsed["mime_type"],
             filename=filename,
             **all_extras,
         )

libs/core/tests/unit_tests/messages/block_translators/test_openai.py

@@ -490,6 +490,16 @@ def test_convert_to_openai_data_block() -> None:
     result = convert_to_openai_data_block(block)
     assert result == expected
 
+    # File / base64 with non-PDF MIME (chat/completions)
+    block = {
+        "type": "file",
+        "base64": "<base64 string>",
+        "mime_type": "text/csv",
+        "filename": "test.csv",
+    }
+    with pytest.raises(ValueError, match="only supports application/pdf"):
+        convert_to_openai_data_block(block)
+
     # Image / base64
     block = {
         "type": "image",
@@ -603,3 +613,39 @@ def test_convert_to_openai_data_block() -> None:
     expected = {"type": "input_file", "file_id": "file-abc123"}
     result = convert_to_openai_data_block(block, api="responses")
     assert result == expected
+
+
+def test_convert_to_v1_file_block_preserves_non_pdf_mime_type() -> None:
+    message = HumanMessage(
+        content=[
+            {
+                "type": "file",
+                "file": {
+                    "filename": "sheet.csv",
+                    "file_data": "data:text/csv;base64,aGVsbG8=",
+                },
+            },
+            {
+                "type": "file",
+                "file": {
+                    "file_data": "data:text/plain;base64,aGVsbG8=",
+                },
+            },
+        ]
+    )
+
+    expected: list[types.ContentBlock] = [
+        {
+            "type": "file",
+            "base64": "aGVsbG8=",
+            "mime_type": "text/csv",
+            "extras": {"filename": "sheet.csv"},
+        },
+        {
+            "type": "file",
+            "base64": "aGVsbG8=",
+            "mime_type": "text/plain",
+        },
+    ]
+
+    assert _content_blocks_equal_ignore_id(message.content_blocks, expected)