fix(reasoning): fall back to reasoning_content when nothing follows </think> tag

Fixes #38197. When a reasoning model's entire answer sits inside
<think>...</think> with nothing trailing after the closing tag,
_parse_content_for_reasoning() previously returned an empty string
as content, discarding the model's only real output even though
finish_reason=stop and the call succeeded.

litellm/litellm_core_utils/prompt_templates/common_utils.py

@@ -1627,7 +1627,14 @@ def _parse_content_for_reasoning(
     )
 
     if reasoning_match:
-        return reasoning_match.group(1), reasoning_match.group(2)
+        reasoning_content = reasoning_match.group(1)
+        content = reasoning_match.group(2)
+        if not content.strip():
+            # Model's entire answer was inside the think block with
+            # nothing after it — surface it as content instead of
+            # silently discarding the model's only real output.
+            content = reasoning_content
+        return reasoning_content, content
 
     return None, message_text
 

tests/minimax_bugfix/test_minimax_reasoning_content_bug.py

@@ -0,0 +1,43 @@
+"""
+Repro for GitHub issue #38197: MiniMax-M2.7 returns message.content empty,
+usage all zero, while reasoning_content has valid output, finish_reason=stop.
+
+Root cause (confirmed against source + MiniMax docs, 2026-08-25):
+MiniMax M2.7 (with reasoning_split unset/false, the default) returns its
+answer wrapped as "<think>...</think><rest of answer>" in a single content
+string. litellm.litellm_core_utils.prompt_templates.common_utils.
+_parse_content_for_reasoning() splits this with a regex whose second capture
+group is everything AFTER the closing </think> tag. When the model's real
+answer sits entirely inside the <think> block with nothing after it, that
+capture group is an empty string — so `content` comes back "" while the
+actual answer is sitting, discarded, in `reasoning_content`.
+
+Tests the parsing function directly — no network, no API key, no mock
+server needed, since this isolates exactly where the bug lives.
+"""
+from litellm.litellm_core_utils.prompt_templates.common_utils import (
+    _parse_content_for_reasoning,
+)
+
+
+class TestMinimaxReasoningContentBug:
+    def test_answer_entirely_inside_think_tag_yields_empty_content(self):
+        raw = "<think>The answer to 2+2 is 4.</think>"
+
+        reasoning_content, content = _parse_content_for_reasoning(raw)
+
+        print("reasoning_content:", repr(reasoning_content))
+        print("content:", repr(content))
+
+        assert reasoning_content == "The answer to 2+2 is 4."
+        # Fixed: content now falls back to reasoning_content when nothing
+        # follows the closing </think> tag, instead of being empty.
+        assert content == "The answer to 2+2 is 4."
+
+    def test_answer_after_think_tag_still_works(self):
+        raw = "<think>Let me work this out.</think>The answer is 4."
+
+        reasoning_content, content = _parse_content_for_reasoning(raw)
+
+        assert reasoning_content == "Let me work this out."
+        assert content == "The answer is 4."