diff --git a/studio/backend/routes/inference.py b/studio/backend/routes/inference.py
index 99910808f8..40e35cc0f6 100644
--- a/studio/backend/routes/inference.py
+++ b/studio/backend/routes/inference.py
@@ -1200,11 +1200,18 @@ async def openai_chat_completions(
 
                         # "content" type -- cumulative text
                         cumulative = event.get("text", "")
-                        # Strip tool-call XML that may have leaked
-                        # through the backend's content stream.
-                        for pat in _tool_xml_strip:
-                            cumulative = pat.sub("", cumulative)
-                        cumulative = cumulative.rstrip()
+                        # Strip closed tool-call XML pairs that may
+                        # have leaked through the backend stream.
+                        # Only strip closed pairs here (not open-ended)
+                        # so legitimate text after a tool block is kept.
+                        cumulative = _re.sub(
+                            r"<tool_call>.*?</tool_call>", "",
+                            cumulative, flags = _re.DOTALL,
+                        )
+                        cumulative = _re.sub(
+                            r"<function=\w+>.*?</function>", "",
+                            cumulative, flags = _re.DOTALL,
+                        )
                         new_text = cumulative[len(prev_text) :]
                         prev_text = cumulative
                         if not new_text: